No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
An Investigation of the Modified Direction Feature for Cursive Character Recognition
Abstract
This paper describes and analyses the performance of a novel feature extraction technique for the recognition of segmented/cursive characters that may be used in the context of a segmentation-based handwritten word recognition system. The modified direction feature (MDF) extraction technique builds upon the direction feature (DF) technique proposed previously that extracts direction information from the structure of character contours. This principal was extended so that the direction information is integrated with a technique for detecting transitions between background and foreground pixels in the character image. In order to improve on the DF extraction technique, a number of modifications were undertaken. With a view to describe the character contour more effectively, a re-design of the direction number determination technique was performed. Also, an additional global feature was introduced to improve the recognition accuracy for those characters that were most frequently confused with patterns of similar appearance. MDF was tested using a neural network-based classifier and compared to the DF and transition feature (TF) extraction techniques. MDF outperformed both DF and TF techniques using a benchmark dataset and compared favourably with the top results in the literature. A recognition accuracy of above 89% is reported on characters from the CEDAR dataset. Yes Yes
... Setelah pixel foreground yang bernilai 1 diberi nilai sesuai gesture arah ketetanggaan, selanjutnya akan dihitung posisi dan jumlah transisi dari suatu piksel citra dengan mengggunakan rumus pencaian nilai LT dan DT. Gambar 10 merupakan contoh ilusrasi perhitungan nilai LT dan DT dari arah kiri ke kanan pada posisi [5,10], [5,11], [11,11], [5,12], dan [12,12]. Dimana adalah indeks pixel yang dikaji dihitung dari awal pencarian dan merupakan jumlah pixel maksimal dalam satu baris atau kolom segmen pixel citra. ...
... Setelah pixel foreground yang bernilai 1 diberi nilai sesuai gesture arah ketetanggaan, selanjutnya akan dihitung posisi dan jumlah transisi dari suatu piksel citra dengan mengggunakan rumus pencaian nilai LT dan DT. Gambar 10 merupakan contoh ilusrasi perhitungan nilai LT dan DT dari arah kiri ke kanan pada posisi [5,10], [5,11], [11,11], [5,12], dan [12,12]. Dimana adalah indeks pixel yang dikaji dihitung dari awal pencarian dan merupakan jumlah pixel maksimal dalam satu baris atau kolom segmen pixel citra. ...
... Setelah pixel foreground yang bernilai 1 diberi nilai sesuai gesture arah ketetanggaan, selanjutnya akan dihitung posisi dan jumlah transisi dari suatu piksel citra dengan mengggunakan rumus pencaian nilai LT dan DT. Gambar 10 merupakan contoh ilusrasi perhitungan nilai LT dan DT dari arah kiri ke kanan pada posisi [5,10], [5,11], [11,11], [5,12], dan [12,12]. Dimana adalah indeks pixel yang dikaji dihitung dari awal pencarian dan merupakan jumlah pixel maksimal dalam satu baris atau kolom segmen pixel citra. ...
... Analysis on contours allows us to compensate from the sensitivity of the writing instrument. Contours are expressed by a sequence of chain codes, which have been widely applied to character recognition [7] and writer recognition [23] . Because the work of gender identification also uses handwriting images, we applied the chain code representation for feature extraction. ...
... The gender of the remaining 193 writers is required to predict. [7] 40 Geometrical Yes ICDAR 2013 Curvature [23] 900 Geometrical Yes ICDAR 2013 Chain Code [22] 5020 Geometrical Yes ICDAR 2013 Gradient Direction [16] 1096 Transformed Yes ICDAR 2013 Fourier [9] 1600 Transformed Yes RDF Gabor [28] 36 Transformed Yes RDF GMSF [15] 6561 Geometrical Yes RDF Allographic [8] 1296 Geometrical Yes RDF Shape + Structure [25] 24 Geometrical No RDF Our registration-form documents (RDF) set has 11,118 handwritten documents which consist of 9256 gray images and 1862 color images. In this paper, we consider each type of field in the registration forms except for the sheet letters, which were machineprinted. ...
... Accuracy for this type of feature was 59.2%. The performance of the Direction [7] features had a high accuracy at 61.7%. Classification accuracy of Curvature [23] was lowest, indicating that the Direction [7] features were better than the Curvature [23] features. ...
This paper presents a new flexible approach to predict the gender of the writers from their handwriting samples. Handwriting features like slant, curvature, line separation, chain code, character shapes, and more, can be extracted from different methods. Therefore, the multi-feature sets are irrelevant and redundant. The conflict of the features exists in the sets, which affects the accuracy of classification and the computing cost. This paper proposes an approach, named kernel mutual information (KMI), that focuses on feature selection. The KMI approach can decrease redundancies and conflicts. In addition, it extracts an optimal subset of features from the writing samples produced by male and female writers. To ensure that KMI can apply the various features, this paper describes the handwriting segmentation and handwritten text recognition technology used. The classification is carried out using a Support Vector Machine (SVM) on two databases. The first database comes from the ICDAR 2013 competition on gender prediction, which provides the samples in both Arabic and English. The other database contains the Registration-Document-Form (RDF) database in Chinese. The proposed and compared methods were evaluated on both databases. Results from the methods highlight the importance of feature selection for gender prediction from handwriting.
... The purpose of feature extraction is to achieve most relevant and discriminative features to identify a symbol uniquely (Blumenstein et al. 2007). Many feature extraction technique are proposed and investigated in the literature that may be used for numeral and character recognition. ...
... Many feature extraction technique are proposed and investigated in the literature that may be used for numeral and character recognition. Consequently, recent techniques show very promising results for separated handwritten numerals recognition (Wang et al. 2005), however the same accuracy has not been attained for cursive character classification (Blumenstein et al. 2007). It is mainly due to ambiguity of the character without context of the entire word (Cavalin et al. 2006). ...
... Recently, neural network classifiers are proved to be powerful and successful for character/word recognition (Verma et al. 2004;Blumenstein et al. 2007). However, to improve the intelligence of these ANNs, huge iterations, complex computations, and learning algorithms are needed, which also lead to consume the processor time. ...
The choice of pattern classifier and the technique used to extract the features are the main factors to judge the recognition accuracy and the capability of an Optical Character Recognition (OCR) system. The main focus of this work is to extract features obtained by binarization technique for recognition of handwritten characters of English language. The recognition of handwritten character images have been done by using multi-layered feed forward artificial neural network as a classifier. Some preprocessing techniques such as thinning, foreground and background noise removal, cropping and size normalization etc. are also employed to preprocess the character images before their classification. Very promising results are achieved when binarization features and the multilayer feed forward neural network classifier is used to recognize the off-line cursive handwritten characters.
... The literature is abounded with highly accurate systems for segmentation and recognition of handwritten numerals [11,12]. Indeed, latest techniques demonstrated potential results [13], but they could not achieve the same accuracy rate for cursive and segmented characters [14][15][16][17][18][19][20][21][22]. Although many problems occur during the recognition of segmented handwritten character, however, there are three main issues: firstly, the first problem concerns with the vagueness of a character without the perspective of the intact word. ...
... For example, "l" is quite similar to "i". Secondly, the problem occurs because of cursive writing style which has ornaments and imprecise character shape that subsequently make the certain characters illegible [21]. ...
... Additionally, researchers employ learning vector quantization (LVQ), neural gas and LVQ classifiers' combination for CEDAR cursive characters' recognition. Recently, Blumenstein et al. [21] investigate an improved feature extraction technique that concerns with local features and in this technique, character contours provide direction information. In this anticipated technique, the foreground pixels are replaced from a character contour to suitable numerical direction values using chain code information. ...
This paper presents an improved feature extraction technique for the cursive characters recognition. This technique can be applied in the perspective of handwritten word recognition system based on segmentation. The bases of fused statistical features extraction technique are improved projection profile and transition features. To extend this principal, a technique is integrated with the projection profile information to detect shifts of background and foreground pixels in the image of a character. A classifier based on neural network is used to test the improved fused features and comparison is done with the projection profile (PP) and transition feature (TF) extraction techniques. By using standard dataset, PP and TF techniques altogether show best performance with fused features having new enhancements and the best results in the literature are compared promisingly with this technique. The characters that are taken from the CEDAR dataset show 91.38% recognition accuracy.
... The purpose of feature extraction is to achieve most relevant and discriminative features to identify a symbol uniquely (Blumenstein et al. 2007). In OCR applications, extracted features used to distinguish between all existing character classes. ...
... Accordingly, many feature extraction technique are proposed and investigated in the literature that may be used for numeral and character recognition. Consequently, recent techniques show very promising results for separated handwritten numerals recognition (Wang et al. 2005), however the same accuracy has not been attained for cursive character classification (Blumenstein et al. 2007). It is mainly due to ambiguity of the character without context of the entire word (Cavalin et al. 2006). ...
... According to Suen (1986), there are two main categories of features: statistical features and structure features. Statistical features derive from statistical distribution of every point in a character matrix such as moments, histograms, profile projection and zoning (Kimura et al. 1992;Blumenstein et al. 2003Blumenstein et al. , 2007Kim et al. 2000;Vamvakas et al. 2007). ...
This paper presents detailed review in the field of off-line cursive script recognition. Various methods are analyzed that
have been proposed to realize the core of script recognition in a word recognition system. These methods are discussed in
view of the two most important properties of such systems: size and nature of the lexicon involved and whether or not a segmentation
stage is present. Script recognition techniques are classified into three categories: firstly, segmentation-free methods or
holistic approaches, that compare a sequence of observations derived from whole word image with similar references of words
in the small lexicon. Secondly, segmentation-based methods, that look for the best match between consecutive sequences of
primitive segments and letters of a possible word similar to human-like reading technique, in which secure features found
all over the word are used to boot-strap a few candidates for a final evaluation phase; thirdly, hybrid approaches. Additionally,
different feature extraction techniques are elaborated in conjunction with the classification process. In this scenario, implications
of single and multiple classifiers are also observed. Finally, remaining problems are highlighted along with possible suggestion
and strategies to solve them.
KeywordsScript recognition–Character segmentation–Character recognition–Feature extraction–Holistic approaches
... To determine the best segmentations, many research works studied merging segments of the word image and invoking a classifier to score the combinations. Most techniques employ an optimization algorithm making use of some sort of lexicon-driven and dynamic programming techniques [22]. ...
... Then, the directions of line segments comprising the characters are detected and the foreground pixels are replaced with appropriate direction values. Finally, features of the characters, based on the location of background to foreground pixel transitions, are extracted and neural training and classification is performed [22,25]. ...
... Blumenstein et al. [22,25] facilitated the extraction of direction features, the following steps are used to prepare the character pattern: ...
Off-line Arabic handwriting recognition and segmentation has been a popular field of research for many years. It still remains an open problem. The challenging nature of handwriting recognition and segmentation has attracted the attention of researchers from industry and academic circles. Recognition and segmentation of Arabic handwritten script is a difficult task because the Arabic handwritten characters are naturally both cursive and unconstrained. The analysis of Arabic script is more complicated in comparison with English script. It is believed, good segmentation is one reason for high accuracy character recognition. This paper proposes and investigates four main segmentation techniques. First, a new feature-based Arabic heuristic segmentation AHS technique is proposed for the purpose of partitioning Arabic handwritten words into primitives (over-segmentations) that may then be processed further to provide the best segmentation. Second, a new feature extraction technique (modified direction features—MDF) with modifications in accordant with the characteristics of Arabic scripts is also investigated for the purpose of segmented character classification. Third, a novel neural-based technique for validating prospective segmentation points of Arabic handwriting is proposed and investigated based on direction features. In particular, the vital process of handwriting segmentation is examined in great detail. The classifier chosen for segmentation point validation is a feed-forward neural network trained with the back-propagation algorithm. Many experiments were performed, and their elapsed CPU times and accuracies were reported. Fourth, new fusion equations are proposed and investigation to examine and evaluate a prospective segmentation points by obtaining a fused value from three neural confidence values obtained from right and center character recognition outputs in addition to the segmentation point validation (SPV) output. Confidence values are assigned to each segmentation point located through feature detection. All techniques components are tested on a local benchmark database. High segmentation accuracy is reported in this research along with comparable results for character recognition and segmentation.
... Figure 27 shows samples of the ACDAR dataset. The results from the proposed SLDF1 technique were comparatively evaluated with results from both techniques by Al Hamad and Abu Zitar (2010) and Blumenstein et al. (2007). The number of processed pixels needed to accomplish the strokes labelling for all 500 word images of the ACDAR data set were used to determine technique speed using the following equations: ...
... Different from the proposed method, the Al Hamad and Zitar's method detects segmentation points based on the text skeleton and modified vertical projection histogram, which leads to producing more candidate segmentation points, which are validated via a set of direction features and neural networks. Based on the measurement criteria results of the techniques of Blumenstein et al. (2007), Al Hamad and Abu Zitar (2010), and our proposed SLDF1, it is clear that the proposed SLDF1 technique is superior in performance as follows: ...
Offline Character segmentation of text images is an important step in many document image analysis and recognition (DIAR) applications. However, the character segmentation of both writing styles (printed and handwritten) remains an open problem. Moreover, the manual segmentation is time-consuming and impractical for large numbers of documents. Based on the unconstraint-cursive handwritten perspective, the auto character segmentation is more challenging and complex. The Arabic script writing style suffers from many common problems, such as sub-words overlapping, characters overlapping, and missed characters. These challenging issues have attracted the attention of researchers in the field of DIAR for Arabic character segmentation. The proposed method combines a new advanced Stroke Labelling based on Direction Features (SLDF2) technique and a modified vertical projection histogram (MVPH) technique. This technique extracts the relationship between each text stroke pixel and its 8 neighboring foreground pixels and labels it with the proper value before identify the possible segmentation points. The text preparation for the segmentation process was achieved using multiple preprocessing steps and developing an advanced stroke labelling technique based on direction features. Several Arabic language structural-rules were made to detect the candidate segmentation points (CSP), detect many character overlapping cases, solve the missed characters problem that appears as a result of using the text skeleton in VPH, and validate the CSP. All techniques and methods are tested on the ACDAR benchmark database. The validation method used to measure segmentation accuracy was a quantitative analysis that includes Recall, Precision, and F-measurement tests. The average accuracy of the proposed segmentation method was 92.44%, which outperforms the state-of-the-art method.
... To promote research on machine learning and pattern recognition, several standard databases have emerged. The handwritten digits are preprocessed, including segmentation and normalization, so that researchers can compare recognition results of their techniques on a common basis as well as reduce the workload [6,7]. ...
... Extreme learning machine also preferred an h(A) function that was the network activation function. The resulting matrix is shown in Eq. (6). Equation (7) represents the column vector of the resulting matrix T. ...
A container is a gathering of neurons whose action vector speaks to the instantiation parameters of a particular kind of substance, for example, an item or an article part. We utilize the length of the action vector to speak to the likelihood that the substance exists and its introduction to speak to the instantiation parameters. Compelling cases at one dimension make expectations, utilizing change networks, for the instantiation parameters of more elevated amount containers. At the point when different forecasts concur, a higher amount of container ends up dynamic. We demonstrate that a discriminatively prepared, multilayer case framework accomplishes best in class execution on Modified National Institute of Standards and Technology (MNIST) and is extensively superior to a deep learning algorithm at perceiving exceedingly covering digits. Deep learning algorithms encouraged by the function and structure of the brain. The deep extreme learning machine (DELM) approach is used to construct a compound that has the least error and highest reliability. All layers are jointly or greedily optimized, depending on the strategy. Deep extreme learning learns all the layers. This paper shows research on the expectation of the MNIST dataset using a DELM. In this article to predict digits better, we have used feedforward and backward propagation deep learning neural networks. When the results were considered, it was observed that deep extreme learning neural network has the highest accuracy rate with 70% of training (42,000 samples), 30% of test and validation (28,000 examples). When comparing the results, it was seen that the intelligent routing between capsules empowered with DELM (IRBC DELM) has the highest precision rate of 97.8%. Simulation results validate the prediction effectiveness of the proposed DELM strategy.
... The neuron will then apply an activation function such as the hard limiter, in which it is either assigned +1 or −1 if the value is positive or negative, respectively, or sigmoidal functions, which is displayed in Equation 24.2, to derive a value for the output. With regard to training the network, there are several techniques available such as the back-propagation (BP) [35,36] and genetic algorithms [37,38], which are both considered to be leaders with regard to the configuration of ANN: ...
... BP relies on the concept of training the network by propagating error back through the network via modifying the weights after the output has been calculated [35]. The algorithm uses either a predetermined limited amount of iterations or the root-mean-square (RMS) error threshold of the calculated output as stopping criteria [36]. ...
Although governments worldwide have invested significantly in intelligent sensor network research and applications, few books cover intelligent sensor networks from a machine learning and signal processing perspective. Filling this void, Intelligent Sensor Networks: The Integration of Sensor Networks, Signal Processing and Machine Learning focuses on the close integration of sensing, networking, and smart signal processing via machine learning.
Based on the world-class research of award-winning authors, the book provides a firm grounding in the fundamentals of intelligent sensor networks, including compressive sensing and sampling, distributed signal processing, and intelligent signal learning. Presenting recent research results of world-renowned sensing experts, the book is organized into three parts:
Machine Learning—describes the application of machine learning and other AI principles in sensor network intelligence—covering smart sensor/transducer architecture and data representation for intelligent sensors
Signal Processing—considers the optimization of sensor network performance based on digital signal processing techniques—including cross-layer integration of routing and application-specific signal processing as well as on-board image processing in wireless multimedia sensor networks for intelligent transportation systems
Networking—focuses on network protocol design in order to achieve an intelligent sensor networking—covering energy-efficient opportunistic routing protocols for sensor networking and multi-agent-driven wireless sensor cooperation
Maintaining a focus on "intelligent" designs, the book details signal processing principles in sensor networks. It elaborates on critical platforms for intelligent sensor networks and illustrates key applications—including target tracking, object identification, and structural health monitoring. It also includes a paradigm for validating the extent of spatiotemporal associations among data sources to enhance data cleaning in sensor networks, a sensor stream reduction application, and also considers the use of Kalman filters for attack detection in a water system sensor network that consists of water level sensors and velocity sensors.
... As shown in Fig. 6 It is difficult to compare results in handwriting recognition due to different databases used in the experimentations. That being said, SVM recognition rate (89.01%) compares significatively better with other results 5 [28][29][30][31][32][33][34] on cursive character recognition. Finally, SVM recognition rate is comparable with 4 For MLP are, respectively, 71.42%, 82.56% and 88.60%. ...
... Finally, SVM recognition rate is comparable with 4 For MLP are, respectively, 71.42%, 82.56% and 88.60%. 5 The best result [28] is 89.01%. the result (90.24%) obtained by Liu and Blumenstein [35] on a smaller test (∼ 2000 characters) than the one (∼ 19 000 characters) used in our experiments. ...
This paper presents a cursive character recognizer, a crucial module in any cursive word recognition system based on a segmentation and recognition approach. The character classification is achieved by using support vector machines (SVMs) and a neural gas. The neural gas is used to verify whether lower and upper case version of a certain letter can be joined in a single class or not. Once this is done for every letter, the character recognition is performed by SVMs. A database of 57 293 characters was used to train and test the cursive character recognizer. SVMs compare notably better, in terms of recognition rates, with popular neural classifiers, such as learning vector quantization and multi-layer-perceptron. SVM recognition rate is among the highest presented in the literature for cursive character recognition.
... In traditional automated production measurement, the typical method is to use calipers, goniometers, or angle gauges to measure a certain parameter of the workpiece several times, then the average value [1]. With the rapid development of science and technology and the popularity of machine vision technology, there are more and more algorithms for image detection and recognition, such as the non-contact rotation angle measurement method based on a monocular camera [2], image direction determination based on feature extraction [3], dynamic angle measurement method based on machine vision [4], 3D vertebra rotation angle automatic measurement direction recognition based on deep learning [5]. The traditional machine vision algorithm based on image processing adopted in this paper has the advantages of non-contact [6], high resolution, and robustness to free motion and rotational fluctuations, which makes it easy to get the angle between the U-shaped notch of the iron cold and the actual alignment [7], and then achieve the notch of the iron cold to realize automatic alignment through the rotation of the iron cold driven by the motor. ...
... Writing with orientation either to left or right side is one of the writing styles of people. The features which represent both direction and structural information of online handwritings are discussed in [6]. The Delaunay triangulation based shape feature extraction from the online handwritten data is explained in [7]. ...
... Chain codes have been applied to a number of shape recognition problems with varying degrees of success. In case of document recognition, chain codes have been applied to problems like writer identification and verification [21,31], character and word recognition [32,33,34,35], classification of writing styles [36] and handwriting based characterization of writer demographics [37]. In our case, we exploit the chain code representation of the contour of the word shape to extract local orientation and curvature information. ...
This work presents a language independent keyword based document indexing and retrieval system using SVM as classifier. Word spotting presents an attractive alternative to the traditional Optical Character Recognition (OCR) systems where instead of converting the image into text, retrieval is based on matching the images of words using pattern classification techniques. The proposed technique relies on extracting words from images of handwritten documents and converting each word image into a shape represented by its contour. A set of multiple features is then extracted from each word image and instances of same words are grouped into clusters. These clusters are used to train a multi-class SVM which learns different word classes. The documents to be indexed are segmented into words and the closest cluster for each word is determined using the SVM. An index file is maintained for each word containing the word locations within each document. A query word presented to the system is matched with the clusters in the database and the documents containing occurrences of the query word are retrieved. The system realized promising precision and recall rates on the IAM database of handwritten documents.
... When attempting to configure the Neural Network weights, one method that exists is to utilise training algorithms. Two dominant training algorithms that have been proven to excel in network training are the Back-Propagation Algorithm [11,12] and the Evolutionary Neural Network [13,14]. Back-Propagation relies on the concept of training the network by propagating error back through the network via modifying the weights after the output has been calculated. ...
... Chain codes have been effectively applied to problems like character/word recognition [5,6,24,33], classification of writing styles [31] and writer identification [43,44]. Since our task of gender classification also comprises handwritten documents, we expect chain code representation to be effective for feature extraction. ...
This paper presents a study to predict gender of individuals from scanned images of their handwritings. The proposed methodology is based on extracting a set of features from writing samples of male and female writers and training classifiers to learn to discriminate between the two. Writing attributes like slant, curvature, texture and legibility are estimated by computing local and global features. Classification is carried out using artificial neural networks and support vector machine. The proposed technique evaluated on two databases under a number of scenarios realized interesting results on predicting gender from handwriting.
... They reported an accuracy of 94.73% on CEDAR and 99.03% on MNIST character database. Blumenstein et al [17] proposed feature extraction technique for the recognition of segmented/cursive characters. The modified direction feature (MDF) extraction technique build upon the direction feature (DF) technique that extracts direction information from the structure of character contours. ...
In this work, an attempt is made to extract minimum number of features to represent the pattern used as inputs for Feed Forward Back Propagation Neural Network (FFBPNN). The binary image of a pattern stored in the frame is partitioned into square regions. A feature from each region is computed by the density and co-ordinate distance of 1 s. pixels. The neural network is trained with the extracted features and Root Mean Square Error (RMSE) obtained in the training process is used as performance indicator to stop the FFBPNN learning. Tested the proposed feature extraction and classification algorithms on the handwritten numeral database and found very good classification recognition rate.
... Location transitions (LTs) are similarly calculated for each row and each column in both directions, with the relative start positions of each direction feature calculated as a proportion of the total width (in the case of a row) or height (in the case of a column). Given the initial set of LT and DT values corresponding to the actual number of rows and columns in the original character bitmap, the data is normalised and locally averaged to fit into a space of 5 rows and 5 columns producing a final vector of 120 features [4]. ...
This paper presents an approach to handwriting character recognition using recurrent neural networks. The method Multi-dimensional Recurrent Neural Network is evaluated against the classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined with the original MDRNN in cases of confusion letters to avoid misclassifications. The performance of the method is verified in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results.
... paper), The segmentation [3,4] of Arabic handwritten characters have been an area of great interest in the past few years. On-line handwriting recognition refers to automatically recognizing handwritten characters using real-time information such as pressure and 978·1-4673·1535·7/121$31.00 ©2012 IEEE the order of strokes made by a writer usually employing a stylus and pressure sensitive tablet [5,6]. ...
As is well known, good segmentation is one reason for high accuracy of character recognition; this paper proposes and investigates a new technique for segmentation of handwritten Arabic scripts. A new Arabic heuristic segmenter (AHS) has been implemented. The AHS employs three new features to locate a Prospected Segmentation Point (PSP) based on shape of the word image, first, remove the punctuation marks (dots), second, ligature detection, and third, additional techniques. Remove the punctuation marks technique has been used to avoid the overlap of the ligature to decrease errors of "missed" and "bad" segmentation points. Ligature detection technique has been used to improve locate the segmentation points that calculated based on distance between local minima and maxima of histogram, the technique calculated based on the distance between foreground and background pixels of word image histogram. An additional technique that contains the average character width technique, and close/open holes detection technique has also been investigated to enhance the overall results of segmentation.
... Location transitions (LTs) are similarly calculated for each row and each column in both directions, with the relative start positions of each direction feature calculated as a proportion of the total width (in the case of a row) or height (in the case of a column). Given the initial set of LT and DT values corresponding to the actual number of rows and columns in the original character bitmap, the data is then normalised and locally averaged to fit into a space of 5 rows and 5 columns producing a final vector of 120 features [2]. Camastra [7] presented in this work a cursive character recognizer, a crucial module in any cursive word recognition system based on a segmentation and recognition approach. ...
This paper presents a hybrid KNN-SVM method for cursive character recognition. Specialized Support Vector Machines (SVMs) are introduced to significantly improve the performance of KNN in handwrite recognition. This hybrid approach is based on the observation that when using KNN in the task of handwritten characters recognition, the correct class is almost always one of the two nearest neighbors of the KNN. Specialized local SVMs are introduced to detect the correct class among these two different classification hypotheses. The hybrid KNN-SVM recognizer showed significant improvement in terms of recognition rate compared with MLP, KNN and a hybrid MLP-SVM approach for a task of character recognition.
... The stable states of the network represent the stored patterns. Neural networks are often used for pattern recognition and classification [8][10]. Hopfield (1982) proposed a fully connected neural network model of associative memory in which we can store information by distributing it among neurons, and recall it from the neuron states dynamically relaxed. ...
In the present paper, an effort has been made to compare and analyze the performance for pattern recalling with conventional hebbian learning rule and with evolutionary algorithm in Hopfield Model of feedback Neural Networks. A set of ten objects has been considered as the pattern set. In the Hopfield type of neural networks of associative memory, the weighted code of input patterns provides an auto-associative function in the network. The storing of the objects has been performed using Hebbian rule and recalling of these stored patterns on presentation of prototype input patterns has been made using both - conventional hebbian rule and evolutionary algorithm. Exploration of the population generation techniques (mutation and elitism), crossover and setting up of proper fitness evaluation functions to generate the new population of the weight matrices from the optimal weight matrix of the stored patterns has been done. The simulated results show that the genetic algorithm is the best searching technique to recall the approximate input patterns.
... Genetic algorithms and neural networks can be integrated into a single application to take advantage of the best features of these technologies [22]. Much work has been done on the evolution of neural networks with GA [23][24][25][26][27]. There have been a lot of researches which apply evolutionary techniques to layered neural networks. ...
In this paper we are studying the performance of Hopfield neural network for recalling of memorized patterns from the Hebbian rule and genetic algorithm for English characters. In this process the genetic algorithm is employed in random form and sub-optimal form for recalling of memorized patterns corresponding to the presented noisy prototype input patterns. The objective of this study is to determine the optimal weight matrix for correct recalling corresponding to noisy form of the English characters. In this study the performance of neural network is evaluated in terms of the rate of success for recalling of noisy input patterns of the English characters with GA in two aspects. The first aspect reflects the random nature of the GA and the second one exhibits the suboptimal nature of the GA for its exploration. The simulated results demonstrate the better performance of network for recalling of the stored letters of English alphabets using genetic algorithm on the suboptimal weight matrix.
... Promising results reported to confirm the best combination of DDD feature and the map tiling. Blumenstein et al. (2004Blumenstein et al. ( , 2007 and Verma et al. (2004) use directional features extracted from character contours. The technique replaces foreground pixels of character contours with suitable direction values. ...
Neural network are most popular in the research community due to its generalization abilities. Additionally, it has been successfully implemented in biometrics, features selection, object tracking, document image preprocessing and classification. This paper specifically, clusters, summarize, interpret and evaluate neural networks in document Image preprocessing. The importance of the learning algorithms in neural networks training and testing for preprocessing is also highlighted. Finally, a critical analysis on the reviewed approaches and the future research guidelines in the field are suggested.
... Therefore, for each genuine signature there are 30 skilled forgeries made by 10 forgers from 10 different genuine specimens. After the modified direction feature (MDF) [28] is used to extract features, we employ PCA to reduce dimensions. The dimension is reduced from 2400 to 113. ...
... Feature extraction method includes Template matching, Deformable templates, Unitary image transforms, Graph description, Projection histograms, Contour profiles, Zoning, Geometric moment invariants, Zernike moments, Spline curve approximation and Fourier descriptors. Different methods like neural network [7,8], Support vector machines [9], Fuzzy logic based [10] HCR are reported for the recognition of ...
In this paper, we propose unconstrained handwritten Kannada character recognition based on Fisher Linear Discriminant Analysis (FLD). The proposed system extracts features from well known FLD, Two dimensional FLD (2D-FLD) and Diagonal FLD. In order to classify the characters, we explore different distance measure techniques and compare their results. The proposed system is tested on unconstrained handwritten Kannada characters with pertaining to large number of character classes. The system showed effectiveness and feasibility of the proposed method.
... Chain codes have shown effective performance on problems like shape registration [42] and object recognition [43]. In case of handwritten document images, these directional features and their improved variants have been successfully applied to character/word recognition [44][45][46][47] as well as classification of writing styles [48]. Since the handwritten characters issued by a particular writer can be regarded as having a specific shape/style, chain code based features are likely to work well on tasks like writer recognition as well. ...
We propose an effective method for automatic writer recognition from unconstrained handwritten text images. Our method relies on two different aspects of writing: the presence of redundant patterns in the writing and its visual attributes. Analyzing small writing fragments, we seek to extract the patterns that an individual employs frequently as he writes. We also exploit two important visual attributes of writing, orientation and curvature, by computing a set of features from writing samples at different levels of observation. Finally we combine the two facets of handwriting to characterize the writer of a handwritten sample. The proposed methodology evaluated on two different data sets exhibits promising results on writer identification and verification.
The world started to talk about optical character recognition (OCR) around 1870. Then over another 25 years OCR systems were designed for industrial applications. And now the OCR software is easily available online for free, through products like Acrobat reader, WebOCR, etc. But still the research is on. Do we need to switch direction or introduce new hypothesis are some of the key questions? The purpose of this chapter is to answer the above questions and propose new methods for character recognition.
Digital forensics has a vital effect in several domains and mainly focuses on reactive measures, especially when facing digital incidents. Gender identification becomes the important problem in the realm of forensic techniques and handwriting recognition. In this paper, attention-based two-pathway Densely Connected Convolutional Networks (ATP-DenseNet) is proposed to identify the gender of handwriting. There are two pathways in ATP-DenseNet: Feature pyramid could extract hierarchical page feature, and attention-based DenseNet (A-DenseNet) could extract the word feature by fusing Convolutional Block Attention Module (CBAM) and dense connected block. Finally, ATP-DenseNet makes the final prediction combining the two pathways. Experimental results show the efficiency of ATP-DenseNet, and the proposed method performs better than other researches. And the visualization of the feature maps can help us to know which part of the image contributes most to the gender identity.
The cursive nature of multilingual characters segmentation and recognition of Arabic, Persian, Urdu languages have attracted researchers from academia and industry. However, despite several decades of research, still multilingual characters classification accuracy is not up to the mark. This paper presents an automated approach for multilingual characters segmentation and recognition. The proposed methodology explores character based on their geometric features. However, due to uncertainty and without dictionary support few characters are over-divided. To expand the productivity of the proposed methodology a BPN is prepared with countless division focuses for cursive multilingual characters. Prepared BPN separates off base portioned indicates effectively with rapid upgrade character acknowledgment precision. For reasonable examination, only benchmark dataset is utilized.
Memetic algorithms (MAs) are originally optimization algorithms with separate individual improvement, and they tend to fully exploit the problem area under consideration. But just like human brain, the recognition time tends to increase with increasing size of population. This paper aims to provide a logical solution using cultural evolution and local learning feature of MA. By introducing best bound population (BBP) from available set of population size, it is possible to keep recognition time in acceptable limits. The best bound population can be continuously upgraded using local search. The paper also revisits some popular techniques of character recognition using traditional approach and using genetic approach. Finally, all techniques are compared for error percentage and recognition time. The relative comparison with figures is presented to justify the findings.
Bu çalışmada, hazırlanan arayüzdeki sınırlandırılmış alana fare ile çizilen Osmanlıca Harflerin tanınması ve Osmanlıca Metin Editörüne aktarılması sağlanmıştır. Arayüzün tasarlanması için Delphi 6.0 Görsel Programlama Dili kullanılmıştır. Tanınan harfler metin editörüne Unicode Karakter Sisteminde aktarılmıştır. Hazırlanan programda Yapısal Yaklaşıma ait Yön Kodlaması Yöntemi kullanılmıştır. Harflerin genel doğru tanınma ortalaması %92’dir. Bu çalışmada öznitelik tabloları Nesih Hattına göre hazırlandığı için, bu hat üzerinde doğru sonuçlara ulaşılmaktadır. Fakat öznitelik tabloları diğer hatlara göre oluşturulduğunda bunların da tanınması sağlanabilmektedir. Tanınan karakterler, Osmanlıca olarak metin editörüne aktarılarak düzenlenebilecek hale getirilmiştir.
The aim of this study is online recognition of Ottoman Letters that drawn on program interface by a mouse and transferring to Ottoman Text Editor. It was prepared with Delphi 6.0 Programming Language in the design of interface. The recognized letters were transferred to Ottoman Text Editor in Unicode Character Set. Direction Coding Method Related to Structural Approach was used in the program. General recognition rate of Ottoman Letters is 92%. In this work, as the feature tables are prepared according to Naskh Fonts, the correct results could be found on this font. If the feature tables are designed according to the other fonts, these fonts can also be recognized. Recognized letters are transferred to Ottoman Text Editor in an editable form.
This paper proposes a word detecting method for document image using character models and word models to evaluate the features of single-character and between-character. First, the text line is segmented into several fragments. Second, the candidate character, which is generated by merging some consecutive fragments, will be identified to be the right one if it conforms to the query word character models. Third, the path search strategy is used to search the candidate words constructed with candidate characters. The word model is used to identify the matching cost. Our experimental results on a dataset of document images demonstrate the effectiveness of the proposed method.
Radio Frequency Identification (RFID) refers to wireless technology that is used to seamlessly and automatically track various amounts of items around an environment. This technology has the potential to improve the efficiency and effectiveness of tasks such as shopping and inventory saving commercial organisations both time and money. Unfortunately, the wide scale adoption of RFID systems have been hindered due to issues such as false-negative and false-positive anomalies that lower the integrity of captured data. In this chapter, we propose the utilisation three highly intelligent classifiers, specifically a Bayesian Network, Neural Network and Non-Monotonic Reasoning, to handle missing, wrong and duplicate observations. After discovering the potential from using Bayesian Networks, Neural Networks and Non-Monotonic Reasoning to correct captured data, we decided to improve upon the original approach by combining the three methodologies into an integrated classifier. From our experimental evaluation, we have shown the high results obtained from cleaning both false-negative and false-positive anomalies using each of our concepts, and the potential it holds to enhance physical RFID systems.
The LCD touch panels have become the basic parts of most products as the human-machine interfaces. The LCD touch panels can become the main parts since their working areas become the bigger, and their real-time properties become the better. That is to say, we can immediately use of the LCD touch panels as the key operating parts of some system for some picture information. It has been known the handwriting touch traces on the LCD touch panel can be displayed immediately on the liquid crystal display, which has the properties like the paper. Therefore, we can use the LCD touch panels as the tool of the handwriting exercising as the paper did. So, we present and study the electronic handwriting exercising technologies (EHET) for Chinese character. The technologies presented consisted of the hardware design, the software design, and the algorithms of the function which were the key of the system designs. Its fundamental operation frames were the page structure. It became a kind of two-part architecture which included a MCU and an integrated operation parts. Its characteristics come from permitting of LCD touch panel and need of the operation task, and made the application wider. The physical prototype implemented proved the above studies mentioned were practical and effective.
This paper proposes a Writer Identification scheme for Malayalam handwritten documents. The novelty of the scheme lies in the fact that the graphemes were used in the training and identification phase of the system. Graphemes are small writing fragments extracted from the handwritten documents which contain meaningful patterns and possess individuality of each writer. The scheme has been tested on a test bed of 280 writers of which 50 writers having only one page, 215 writers with at least 2 pages and 15 writers with at least 4 pages. A recognition rate of 89.28% was achieved.
The problem of handwritten signature recognition is considered significant in biometrics, in particular for determining the validity of official documents. The rationale consists of creating an off-line classifier to discriminate between fake (forged) and genuine digitalized signatures. In such applications containing thousands of samples machine learning techniques such as Support Vector Machines (SVM) play a preponderant role in overcoming the challenges inherent to this problematic. However, to deal with the computational burden of calculating the large Gram matrix, approaches such as Graphics Processing Units (GPU) computing are required for efficiently processing big image biometric data. In this paper, first, we present an empirical study for efficient feature selection concerning the signature identification problem. Second, an GPU-based SVM classifier that integrates a component of the open source Machine Learning Library (GPUMLib) supporting several kernels is developed. Third, we ran several experiments with improved performance over baseline approaches. From our study, we gain insights in both performance and computational cost under a number of experimental conditions, and conclude that the most appropriate model is usually a trade-off between performance and computational cost for a given experimental setup and dataset.
Le travail présenté dans ce manuscrit se situe dans le domaine de l'analyse et la reconnaissance de documents, et plus précisément, la reconnaissance hors-ligne des individus et de leur genre à partir de leur écriture manuscrite. Deux contributions se dégagent de cette étude, dans la première contribution, nous proposons une méthode indépendante du texte pour la reconnaissance du scripteur dans un environnement multi-scripts. L'objectif est de reconnaître l'auteur d'un texte manuscrit dans un script à partir d'échantillons du même auteur dans un autre script et donc valider l'hypothèse que le style d'écriture d'un individu reste constant à travers différents scripts. La méthode proposée est basée sur des distributions de longueurs de segments qui sont comparées avec les méthodes les plus connues et les plus performantes de l’état de l’art. La classification est réalisée en utilisant les K plus proches voisins (k-PPV) ainsi que les séparateurs à vaste marge (SVM). Les résultats expérimentaux obtenus sur une base de données de 126 scripteurs avec 4 échantillons par scripteur montrent que la méthode proposée permet d'obtenir des performances intéressantes. Notre deuxième contribution consiste à présenter une étude pour la détermination du sexe des individus à partir de leurs écritures manuscrites. La méthode proposée est basée sur l'extraction d'un ensemble de caractéristiques de l'écriture à partir d’échantillons de scripteurs de sexe masculin et féminin et l’entraînement d’un classifieur afin qu'il puisse distinguer entre les deux catégories. Des attributs de l'écriture comme l'orientation, la courbure, la texture et la lisibilité sont estimés en calculant des caractéristiques locales et globales. La classification est effectuée à l'aide des réseaux de neurones artificiels (ANN) ainsi que les séparateurs à vaste marge (SVM). La méthode proposée a été évaluée en utilisant deux bases de données sous un certain nombre de scénarios où des résultats intéressants ont été enregistrés.
The aim of this work is to judge the efficiency of Multi Layer Perceptron (MLP) and Radial Basis Function (RBF) neural network classifiers for performing the task of cursive handwritten digit recognition. Binarization features are extracted from the preprocessed handwritten digit images. The features thus obtained are used to train MLP and RBF classifiers. A detailed investigation in the proposed experiment was done and it can be summarized that when binarization features of the digit images are extracted and used for training the neural network classifiers in the recognition experiment, RBF classifier outperforms the MLP classifier. The RBF Network delivers 98.40% recognition accuracy whereas the MLP classifier delivers 96.20% accuracy for the proposed experiment of cursive handwritten digit recognition.
This paper describes the sequence of construction of a cell nuclei classification model by the analysis, the characterization and the classification of shape and texture. We describe first the elaboration of dedicated shape indexes and second the construction of the associated classification submodel. Then we present a new method of texture characterization, based on the construction and the analysis of statistical matrices encoding the texture. The various characterization techniques developed in this paper are systematically compared to previous approaches. In particular, we paid special attention to the results obtained by a versatile classification method using a large range of descriptors dedicated to the characterization of shapes and textures. Finally, the last classifier built with our methods achieved 88% of classification out of the 94% possible.
Research in offline Arabic handwriting recognition has increased considerably in the past few years. This is evident from the numerous research results published recently in major journals and conferences in the area of handwriting recognition. Features and classifications techniques utilized in recent research work have diversified noticeably compared to the past. Moreover, more efforts have been diverted, in last few years, to construct different databases for Arabic handwriting recognition. This article provides a comprehensive survey of recent developments in Arabic handwriting recognition. The article starts with a summary of the characteristics of Arabic text, followed by a general model for an Arabic text recognition system. Then the used databases for Arabic text recognition are discussed. Research works on preprocessing phase, like text representation, baseline detection, line, word, character, and subcharacter segmentation algorithms, are presented. Different feature extraction techniques used in Arabic handwriting recognition are identified and discussed. Different classification approaches, like HMM, ANN, SVM, k-NN, syntactical methods, etc., are discussed in the context of Arabic handwriting recognition. Works on Arabic lexicon construction and spell checking are presented in the postprocessing phase. Several summary tables of published research work are provided for used Arabic text databases and reported results on Arabic character, word, numerals, and text recognition. These tables summarize the features, classifiers, data, and reported recognition accuracy for each technique. Finally, we discuss some future research directions in Arabic handwriting recognition.
This paper presents entry for aura response for character recognition and the handwritten or printed text translation into editable text. The objective is to identify handwritten characters with the help of neural networks and facilitates the conversion of handwritten documents to editable text from document images. Handwritten contentedness boasts challenges that are seldom encountered in machine-printed text. The translation basis is either mechanical or electronic translation. This is not easy since different people have different handwriting styles. Assigning distinct templates to each and every alphabet and numbers is the approach described. This concept can be a trademark in data entry applications. The suggested method is simple, have promising discrimination accuracy and less time complexity.
The studies of Optical Character Recognition (OCR) are being developed since it still needs a performance improvement. The previous study of alphanumeric character recognition had been conducted by Blumenstein and Liu using Modified Direction Feature (MDF) and Multi Layer Perceptrons (MLP) network. The study reaches the accuracy rate of 70.22% for lowercase characters and 80.83% for uppercase characters. In this study the OCR system is proposed to improve the existing performance and have a capability to recognize all case-sensitive alphanumeric characters simultaneously. One of the problems is that there are several characters having similarities in gesture and shape, so that the classifier of the OCR system encounters many ambiguities when classifying some particular characters, especially when recognizing all case-sensitive alphanumeric characters. To overcome those problems, this study proposes a technique of grouping. All character classes are clustered into some groups using Fuzzy C-Means (FCM) clustering method. The Nested MLP is the novelty of the previous method that is implemented in this study. This is a kind of multi-level MLP network that classifies the problem domain hierarchically. The first level classifies the character into the designated group and the second level continues the classification into the recognized character class. The OCR system using the methods to recognize all case-sensitive alphanumeric characters yields an accuracy rate of 84.38% for the uppercases, 76.43% for the lowercases, and 78.92% for the digits respectively. Any misclassified characters are mostly happened in distinguishing several uppercase and lowercase characters having similarities in gestures and shapes.
Data entry forms are employed in all types of enterprises to collect hundreds of customer’s information on daily basis. The information is filled manually by the customers. Hence, it is laborious and time consuming to use human operator to transfer these customers information into computers manually. Additionally, it is expensive and human errors might cause serious flaws. The automatic interpretation of scanned forms has facilitated many real applications from speed and accuracy point of view such as keywords spotting, sorting of postal addresses, script matching and writer identification. This research deals with different strategies to extract customer’s information from these scanned forms, interpretation and classification. Accordingly, extracted information is segmented into characters for their classification and finally stored in the forms of records in databases for their further processing. This paper presents a detailed discussion of these semantic based analysis strategies for forms processing. Finally, new directions are also recommended for future research.
This paper presents a hybrid MLP-SVM method for cursive characters recognition. Specialized Support Vector Machines (SVMs) are introduced to significantly improve the performance of Multilayer Perceptron (MLP) in the local areas around the surfaces of separation between each pair of characters in the space of input patterns. This hybrid architecture is based on the observation that when using MLPs in the task of handwritten characters recognition, the correct class is almost always one of the two maximum outputs of the MLP. The second observation is that most of the errors consist of pairs of classes in which the characters have similarities (e.g. (U, V), (m, n), (O, Q), among others). Specialized local SVMs are introduced to detect the correct class among these two classification hypotheses. The hybrid MLP-SVM recognizer showed improvement, significant, in performance in terms of recognition rate compared with an MLP for a task of character recognition.
As an improved unsupervised learning algorithm of CCA, Kernel Canonical Correlation Analysis (Kernel CCA) can extend CCA to the nonlinear case by applying the kernel trick. In this paper, an optical character recognition system based on image preprocessing technologies combined with Kernel CCA has been developed. Moreover, due to the duality between Kernel CCA and LS-SVM, the optimization problem of Kernel CCA is transformed into the solving of quadratic equations by means of LS-SVM method. The proposed method has been evaluated by carrying out recognition experiments on the optical printed characters of electronic components. The results show that the proposed method has a better recognition performance, and the computational complexity can be simplified largely by introducing LS-SVM method.
This paper proposes to find an optimal feature set for handwritten numeral recognition using the box partitioning method. The feature under study is the mean normalized distance measure which is the most popular descriptor in this regard. However it is always used in combination with other descriptors and does not give good classification results when used on its own. The descriptor vector is obtained by partitioning the numeral image into sub-boxes and computing the distance measure from each sub-box taken in order. A series of evaluations is carried out in this work to verify the optimal size and number of sub-boxes by subjecting the resulting feature vectors to a rigorous handwritten numeral classification test, using a simple MLP neural network classifier. It is proved in our work that better results are obtained when the number of partitions along the horizontal and vertical axis of the image is fixed, rather than the conventional technique of arbitrarily dividing the image into sub-boxes of pre-defined dimensions.
Character Recognition (CR) has been an active area of research in the past and due to its diverse applications it continues to be a challenging research topic. In this paper, we focus especially on offline recognition of handwritten English words by first detecting individual characters. The main approaches for offline handwritten word recognition can be divided into two classes, holistic and segmentation based. The holistic approach is used in recognition of limited size vocabulary where global features extracted from the entire word image are considered. As the size of the vocabulary increases, the complexity of holistic based algorithms also increases and correspondingly the recognition rate decreases rapidly. The segmentation based strategies, on the other hand, employ bottom-up approaches, starting from the stroke or the character level and going towards producing a meaningful word. After segmentation the problem gets reduced to the recognition of simple isolated characters or strokes and hence the system can be employed for unlimited vocabulary. We here adopt segmentation based handwritten word recognition where neural networks are used to identify individual characters. A number of techniques are available for feature extraction and training of CR systems in the literature, each with its own superiorities and weaknesses.We explore these techniques to design an optimal offline handwritten English word recognition system based on character recognition. Post processing technique that uses lexicon is employed to improve the overall recognition accuracy.
Firstly, This paper introduces the application status of the artificial neural network technology in the print character recognition, and then elaborated on the technology of Standard BP neural network. By formula derivation, we showed that Standard BP neural Network exists some defects in the application, and then we take the approach by adding a momentum term to improve the Network, and increases the training speed. Secondly, we randomly selecte 200 printed number-characters and 50 printed letter-characters as a sample of the improved BP neural network experiments, the results show that the method of the number-character recognition rate higher than the alphabetic characters, the performance of convergence speed and recognition is better.
Artificial Neural Networks (ANN) are a classic pattern classifier and widely applicable to various problems and are relatively easy to use. Three of the most popular ANNs are Multilayer Perceptron (MLP) with Backpropagation learning algorithm, Self Organizing Map (SOM) and Recurrent Neural Network (RNN). Support Vector Machines (SVM) have gained great interest in the last few years in pattern recognition. Thus, this research compares the recognition performance of text and non-text images (text, table, figure and graph) from technical document images based on the pixel intensity of various zones between BPNN, SOM, RNN and SVM. Symmetrical and non-symmetrical zoning algorithms were compared as input. 400 different datasets have been tested and the experiments indicate that SVM classification is superior to the other three classifiers. The experiments also indicate that the combination of symmetrical and non-symmetrical zoning design is better than non-symmetrical or symmetrical zoning only.
This paper presents an over-segmentation and validation strategy for off-line cursive handwriting recognition. Over-segmentation module is employed to find all the possible character boundaries. Then, the incorrect segmentation points from over-segmenting module are removed by validating processes. The over-segmentation was performed based on the vertical pixel density between upper and lower baselines. Wherever the pixel density is less than threshold, an over-segmentation point is assigned. After the over-segmentation is done, validation starts removing over-segmentation points. The first validation module checks if a segmentation point lies in hole region. The second validation module compares total foreground pixel between two neighbouring segmentation points to a threshold value. The third validation module is neural network voting by neural network classifier trained on pre-segmented characters. Finally, the oversized segment validation process checks if there is any missing segmentation point between neighbouring characters. The proposed approach has been implemented, and the experiments on CEDAR benchmark database have been conducted. The results of the experiments are very promising and the overall performance of the algorithm is more effective than the other existing segmentation algorithms.
A system is described that performs the following function: given a digital image of a handwritten word and a list of candidates (a lexicon), rank the candidates by how well they match the word image. We refer to this function as word recognition. Word recognition is very important in the interpretation of handwritten address blocks. The described system uses dynamic programming to perform the matching between word images and candidate strings. Neural networks are used to provide character level match strengths. Top choice percentages range from the upper 70% range for lexicons of size 1000 to the mid 80% range for lexicons of size 100 for hand printed words.
Handwriting has continued to persist as a means of communication and recording information in day-to-day life even with the introduction of new technologies. Given its ubiquity in human transactions, machine recognition of handwriting has practical significance, as in reading handwritten notes in a PDA, in postal addresses on envelopes, in amounts in bank checks, in handwritten fields in forms, etc. This overview describes the nature of handwritten language, how it is transduced into electronic data, and the basic concepts behind written language recognition algorithms. Both the online case (which pertains to the availability of trajectory data during writing) and the off-line case (which pertains to scanned images) are considered. Algorithms for preprocessing, character and word recognition, and performance with practical systems are indicated. Other fields of application, like signature verification, writer authentification, handwriting learning tools are also considered
An off-line handwritten word recognition system is described.
Images of handwritten words are matched to lexicons of candidate
strings. A word image is segmented into primitives. The best match
between sequences of unions of primitives and a lexicon string is found
using dynamic programming. Neural networks assign match scores between
characters and segments. Two particularly unique features are that
neural networks assign confidence that pairs of segments are compatible
with character confidence assignments and that this confidence is
integrated into the dynamic programming. Experimental results are
provided on data from the U.S. Postal Service.
A new analytic scheme, which uses a sequence of image segmentation and recognition algorithms, is proposed for the off-line cursive handwriting recognition problem. First, some global parameters, such as slant angle, baselines, stroke width and height, are estimated. Second, a segmentation method finds character segmentation paths by combining gray-scale and binary information. Third, a hidden Markov model (HMM) is employed for shape recognition to label and rank the character candidates. For this purpose, a string of codes is extracted from each segment to represent the character candidates. The estimation of feature space parameters is embedded in the HMM training stage together with the estimation of the HMM model parameters. Finally, information from a lexicon and from the HMM ranks is combined in a graph optimization problem for word-level recognition. This method corrects most of the errors produced by the segmentation and HMM ranking stages by maximizing an information measure in an efficient graph search algorithm. The experiments indicate higher recognition rates compared to the available methods reported in the literature
In this paper, we propose an approach that integrates the
statistical and structural information for unconstrained handwritten
numeral recognition. This approach uses state-duration adapted
transition probability to improve the modeling of state-duration in
conventional HMMs and uses macro-states to overcome the difficulty in
modeling pattern structures by HMMs. The proposed method is superior to
conventional approaches in many aspects. In the statistical and
structural models, the orientations are encoded into discrete codebooks
and the distributions of locations are modeled by joint Gaussian
distribution functions. The experimental results show that the proposed
approach can achieve high performance in terms of speed and accuracy
An image database for handwritten text recognition research is
described. Digital images of approximately 5000 city names, 5000 state
names, 10000 ZIP Codes, and 50000 alphanumeric characters are included.
Each image was scanned from mail in a working post office at 300
pixels/in in 8-bit gray scale on a high-quality flat bed digitizer. The
data were unconstrained for the writer, style, and method of
preparation. These characteristics help overcome the limitations of
earlier databases that contained only isolated characters or were
prepared in a laboratory setting under prescribed circumstances. Also,
the database is divided into explicit training and testing sets to
facilitate the sharing of results among researchers as well as
performance comparisons
In this paper, authors discuss on a lexicon directed algorithm for recognition of unconstrained handwritten words (cursive, discrete, or mixed) such as those encountered in mail pieces. The procedure consists of binarization, pre-segmentation, intermediate feature extraction, segmentation recognition, and post-processing. The segmentation recognition and the post-processing are repeated for all lexicon words while the binarization to the intermediate feature extraction are applied once for an input word. This algorithm is essentially non hierarchical in character segmentation and recognition which are performed in a single segmentation recognition process. The result of performance evaluation using large handwritten address block database, and algorithm improvements are described and discussed to achieve higher recognition accuracy and speed. Experimental studies with about 3000 word images indicate that overall accuracy in the range of 91% to 98% depending on the size of the lexicon (assumed to contain correct word) are achievable with the processing speed of 20 to 30 word per minute on typical work station.
This paper proposes a method for cursive handwritten word recognition. Cursive word recognition generally consists of segmentation of a cursive word, character recognition and word recognition. Traditional approaches detect one candidate of segmentation point between characters, and cut the touching characters at the point [1]. But, it is difficult to detect a correct segmentation point between characters in cursive word, because form of touching characters varies greatly by cases. In this research, we determine multiple candidates as segmentation points between characters. Character recognition and word recognition decide which candidate is the most plausible touching point. As a result of the experiment, at the character recognition stage, recognition rate was 75. 7%, while cumulative recognition rate within best three candidates was 93. 7%. In word recognition, recognition rate was 79. 8%, while cumulative recognition rate within best five candidates was 91. 7% when lexicon size is 50. The processing speed is about 30 sec/word on SPARC station 5.
Discusses improvements made to a lexicon directed algorithm for
recognition of unconstrained handwritten words (cursive, discrete, or
mixed) such as those encountered in mail pieces. The procedure consists
of binarization, pre-segmentation, intermediate feature extraction,
segmentation recognition, and post-processing. The segmentation
recognition and the post-processing are repeated for all lexicon words
while the binarization to the intermediate feature extraction are
applied once for an input word. The result of performance evaluation
using large handwritten address block database is described, and
algorithm improvements are described and discussed, in order to achieve
higher recognition accuracy and speed. As a result the performance for
lexicons of size 10, 100, and 1000 are improved to 98.01%, 95.46%, and
91.49% respectively. The processing speed for each lexicon is improved
to 2.0, 2.5, and 3.5 sec/word on a SUN SPARC station 2
Describes the mixed HMM-KNN word recognition module of a bank
cheque processing system developed at CENPARMI. It uses a combination of
2 segmentation free word recognition schemes. The first scheme uses a
set of global features associated to a modified K nearest neighbour
classifier; while the second one uses a set of directional contour
features as input to an HMM. The system has been designed to be modular
and independent of specific languages as in Canada one has to deal with
at least 2 languages, namely English and French. It can be easily
adapted to read other European languages based on the Roman alphabet.
The system is continuously tested on data from the local phone company,
and we report here the results on a database of approximately 4,500
cheques
This paper presents an innovative approach called box method for feature extraction for the recognition of handwritten characters. In this method, the binary image of the character is partitioned into a fixed number of subimages called boxes. The features consist of vector distance (γ) from each box to a fixed point. To find γ the vector distances of all the pixels, lying in a particular box, from the fixed point are calculated and added up and normalized by the number of pixels within that box. Here, both neural networks and fuzzy logic techniques are used for recognition and recognition rates are found to be around 97 percent using neural networks and 98 percent using fuzzy logic. The methods are independent of font, size and with minor changes in preprocessing, it can be adopted for any language.
This paper gives an assessment of the current state of the art in handwriting recognition. It summarizes the lessons learned, the difficulties involved, and the challenges ahead. Based on a review of the recent achievements in off-line computer recognition of totally unconstrained handwritten characters, and extensive research, the authors attempt to identify new frontiers for research which may lead to further breakthroughs in this field. They will present some evidences and novel ideas on ways of stretching the limits of handwriting recognition systems aiming at outperforming human beings.
In this paper we present LoaFeR, a novel linguistic fuzzy recogniser of off-line isolated handwritten characters. LoaFeR describes the shape of each character by linguistic expressions derived from a fuzzy partition of the character image. A purposely-defined weighted distance is used to compare the linguistic descriptions. A small scale application of LoaFeR, in which 26 lower-case cursive characters written by 60 and 40 different writers were used as training and test sets, respectively, yielded 69.5% recognition rate.
This paper presents a cursive character recognizer, a crucial module in any Cursive Script Recognition system based on a segmentation and recognition approach.The character classification is achieved by combining the use of neural gas (NG) and learning vector quantization (LVQ). NG is used to verify whether lower and upper case version of a certain letter can be joined in a single class or not. Once this is done for every letter, it is possible to find an optimal number of classes maximizing the accuracy of the LVQ classifier.A database of 58000 characters was used to train and test the models. The performance obtained is among the highest presented in the literature for the recognition of cursive characters.
The paper presents a novel approach for extracting structural features from segmented cursive handwriting. The proposed approach is based on the contour code and stroke direction. The contour code feature utilises the rate of change of slope along the contour profile in addition to other properties such as the ascender and descender count, start point and end point. The direction feature identifies individual line segments or strokes from the character's outer boundary or thinned representation and highlights each character's pertinent direction information. Each feature is investigated employing a benchmark database and the experimental results using the proposed contour code based structural feature are very promising. A comparative evaluation with the directional feature and existing transition feature is included.
Handwritten character recognition by human readers, a statistical classifier, and a neural network is compared to know the required accuracy for handwritten word recognition. Sample characters extracted from postal address words on mail pieces collected by USPS were used to evaluate human and machine performance. Experimental results show that: 1) when the characters are segmented from words and are randomly presented, the accuracy of the machine recognition is comparable with the average human recognition accuracy, 2) the neural network employing the feature vector of size 64 outperforms the statistical classifier employing the same feature vector, and that 3) the statistical classifier employing the feature vector of size 400 achieves comparable recognition rate with the best human reader
A new method using Gabor filters for character recognition in gray-scale images is proposed in this paper. Features are extracted directly from gray-scale character images by Gabor filters which are specially designed from statistical information of character structures. An adaptive sigmoid function is applied to the outputs of Gabor filters to achieve better performance on low-quality images. In order to enhance the discriminability of the extracted features, the positive and the negative real parts of the outputs from the Gabor filters are used separately to construct histogram features. Experiments show us that the proposed method has excellent performance on both low-quality machine-printed character recognition and cursive handwritten character recognition.
We describe a method, “Shortest Path Segmentation” (SPS), which combines dynamic programming and a neural net recognizer for segmenting and recognizing character strings. We describe the application of this method to two problems: recognition of handwritten ZIP Codes, and recognition of handwritten words. For the ZIP Codes, we also used the method to automatically segment the images during training: the dynamic programming stage both performs the segmentation and provides inputs and desired outputs to the neural network. Results are reported for a test set of 2642 unsegmented handwritten 212 dpi binary ZIP Code (5- and 9-digit) images.
For handwritten word recognition, we combined SPS with a “Space Displacement Neural Network” approach, in which a single-character-recognition network is extended over the entire word image, and in which SPS techniques are then used to rank order a given lexicon. We report results on a test set of 3000 300 ppi gray scale word images, extracted from images of live mail pieces, for lexicons of size 10, 100, and 1000.
Representing the problem as a graph as proposed in this paper has advantages beyond the efficient finding of the final optimal segmentation, or the automatic segmentation of images during training. We can also easily extend the technique to generate K “runner up” answers (for example, by finding the K shortest paths). This paper will also describe applications of some of these ideas.
This paper describes a neural network-based technique for cursive character recognition applicable to segmentation-based word recognition systems. The proposed research builds on a novel feature extraction technique that extracts direction information from the structure of character contours. This principal is extended so that the direction information is integrated with a technique for detecting transitions between background and foreground pixels in the character image. The proposed technique is compared with the standard direction feature extraction technique, providing promising results using segmented characters from the CEDAR benchmark database.
High accuracy character recognition techniques can provide useful information for segmentation-based handwritten word recognition systems. This research describes neural network-based techniques for segmented character recognition that may be applied to the segmentation and recognition components of an off-line handwritten word recognition system. Two neural architectures along with two different feature extraction techniques were investigated. A novel technique for character feature extraction is discussed and compared with others in the literature. Recognition results above 80% are reported using characters automatically segmented from the CEDAR benchmark database as well as standard CEDAR alphanumerics.
This paper uses a modified Hough transform method to extract features from the cursive handwritten digit and characters in CEDAR database. The technique does not require the detection of complex structural primitives such as loops. The handwriting images are divided into uniform regions that are analysed for the presence of horizontal, vertical and diagonal segments. The total number of such segments found in these regions are used as input to a linear classifier (discriminant analysis) and a nonlinear classifier (nearest neighbour). The results are produced on the complete test sets specified in CEDAR database as well as a leave-one-out cross-validation is performed. On the digit data, the results show a recognition rate of around 94% correct recognition on the test set and 87.5% using the leave-one-out method. Character recognition results range between 67% correct on a test set and 64% connect using leave-one-out cross-validation
Artificial neural networks have been recognized as a powerful tool
for pattern classification problems, but a number of researchers have
also suggested that straightforward neural-network approaches to pattern
recognition are largely inadequate for difficult problems such as
handwritten numeral recognition. In this paper, we present three
sophisticated neural-network classifiers to solve complex pattern
recognition problems: multiple multilayer perceptron (MLP) classifier,
hidden Markov model (HMM)/MLP hybrid classifier, and structure-adaptive
self-organizing map (SOM) classifier. In order to verify the superiority
of the proposed classifiers, experiments were performed with the
unconstrained handwritten numeral database of Concordia University,
Montreal, Canada. The three methods have produced 97.35%, 96.55%, and
96.05% of the recognition rates, respectively, which are better than
those of several previous methods reported in the literature on the same
database
A fast method of handwritten word recognition suitable for real
time applications is presented in this paper. Preprocessing,
segmentation and feature extraction are implemented using a chain code
representation of the word contour. Dynamic matching between characters
of a lexicon entry and segment(s) of the input word image is used to
rank the lexicon entries in order of best match. Variable duration for
each character is defined and used during the matching. Experimental
results prove that our approach using the variable duration outperforms
the method using fixed duration in terms of both accuracy and speed.
Speed of the entire recognition process is about 200 msec on a single
SPARC-10 platform and the recognition accuracy is 96.8 percent are
achieved for lexicon size of 10, on a database of postal words captured
at 212 dpi
In this paper, we propose a new scheme for off-line recognition of
totally unconstrained handwritten numerals using a simple multilayer
cluster neural network trained with the backpropagation algorithm and
show that the use of genetic algorithms avoids the problem of finding
local minima in training the multilayer cluster neural network with
gradient descent technique, and improves the recognition rates. In the
proposed scheme, Kirsch masks are adopted for extracting feature vectors
and a three-layer cluster neural network with five independent
subnetworks is developed for classifying similar numerals efficiently.
In order to verify the performance of the proposed multilayer cluster
neural network, experiments with handwritten numeral database of
Concordia University of Canada, that of Electro-Technical Laboratory of
Japan, and that of Electronics and Telecommunications Research Institute
of Korea were performed. For the case of determining the initial weights
using a genetic algorithm, 97.10%, 99.12%, and 99.40% correct
recognition rates were obtained, respectively
This paper uses a modified Hough transform method to extract features from the cursive handwritten digit and characters CEDAR data. The technique does not require the detection of complex structural primitives such as loops etc. The handwriting images are divided into uniform regions that are analysed for the presence of horizontal, vertical and diagonal segments. The total number of such segments found in these regions are used as input to a linear classifier (discriminant analysis) and a non-linear classifier (nearest neighbour). The results are produced on the complete test sets specified in CEDAR database as well as a leave-one-out crossvalidation is performed. On the digit data, the results show a recognition rate of around 94% correct recognition on the test set and 87.5% using a leave-oneout method. Character recognition results range between 67% correct on test set and 64% correct using leave-oneout cross-validation. 1. Introduction Optical Character Recognition deals with th...
This paper presents a cursive character recognizer embedded in an o-line cursive script recognition system. The recognizer is composed of two modules: the rst one is a feature extractor, the second one an LVQ. The selected feature set was compared to Zernike polynomials using the same classier. Experiments are reported on a database of about 49000 isolated characters. 2 IDIAP{RR 00-47 1
Context-directed handwritten word recognition for postal service applications
- F Kimura
- S Tsuruoka
- M Shridhar
- Z Chen
F. Kimura, S. Tsuruoka, M. Shridhar, Z. Chen, Context-directed handwritten word recognition for postal service applications, Proceedings of the Fifth USPS Advanced Technical Conference, 1992, pp. 199–213.
Design and implementation of a system for recognition of handwritten responses on US census form
- T Bruel
Segmentation based handwritten word recognition
- P D Gader
- M Magdi
- J H Chiang