Figure - uploaded by Oyeranmi Adigun
Content may be subject to copyright.
Source publication
ABSTRACT
Character Recognition has become an intensive research areas
during the last few decades because of its potential
applications. However, most existing classifiers used in
recognizing handwritten online characters suffer from poor
feature selection and slow convergence which affect training
time and recognition accuracy. This paper proposed...
Context in source publication
Context 1
... research combines structural and statistical features to complement each other, handle style variations and distinguish one input pattern from another pattern to achieve better performance. Hence, a hybrid was developed to highlight different character properties that effectively identity a character as shown in Figure 3. Table 2.1 ...Similar publications
![Data trend estimation using MTD function [11]](publication/362590648/figure/fig1/AS:11431281084642514@1663293526505/Data-trend-estimation-using-MTD-function-11_Q320.jpg)
This paper proposes a modified mega trend diffusion (MTD) function based on the K-means clustering algorithm to generate artificial samples for a training dataset. This would improve the prediction accuracy in a backpropagation neural network (BPNN) algorithm used in small dataset problems. The main contribution of this paper is in solving the attr...
In deep learning of classifiers, the cost function usually takes the form of a combination of SoftMax and CrossEntropy functions. The SoftMax unit transforms the scores predicted by the model network into assessments of the degree (probabilities) of an object’s membership to a given class. On the other hand, CrossEntropy measures the divergence of...
Unitary learning is a backpropagation that serves to unitary weights update in deep complex-valued neural network with full connections, meeting a physical unitary prior in diffractive deep neural network ([DN]2). However, the square matrix property of unitary weights induces that the function signal has a limited dimension that could not generaliz...
With the prosperity of mobile devices, the distributed learning approach enabling model training with decentralized data has attracted wide research. However, the lack of training capability for edge devices significantly limits the energy efficiency of distributed learning in real life. This paper describes a novel approach of training DNNs exploi...
Accurate prediction of reservoir landslide displacements is crucial for early warning and hazard prevention. Current machine learning (ML) paradigms for predicting landslide displacement demonstrate superior performance, while often relying on various feature engineering techniques, such as decomposing into different temporal lags and feature selec...
Citations
... Character recognition is an ongoing research which has motivated researchers from various aspect of human endeavors such as image processing, computer vision and machine learning [1]. Online character recognition system is the transformation process, that will allow extraction of input characters, from character image database to digitalize and translates the handwritten text into a machine editable form [2]. Due to global security threat, person authentication, and retrieving text, there is need of adopting techniques that could enhance the recognition performance of the system. ...
... In this article, the first step is the character images were acquired using a pen digitizer from datasets of 6200 samples collected by (Adigun et al., 2016) was used for training. Three preprocessing techniques were employed: binarization, extreme coordinate measurement and grid resizing were used to convert into binary form, measure extreme coordinate of the space and matrix standard respectively. ...
This paper carries out performance evaluation of a Modified Genetic Algorithm (MGA) and Modified Counter Propagation Network (MCPN) Network for online character recognition. Two techniques were used to train the feature vectors using supervised and unsupervised methods. First, the modified genetic algorithm used feature selection to filter irrelevant features vectors and improve character recognition because of its stochastic nature. Second, MCPN has its capability to extract statistical properties of the input data. MGA and MCPN techniques were used as classifiers for online character recognition then evaluated using 6200-character images for training and testing with best selected similarity threshold value. The experimental results of evaluation showed that, at 5 x 7 pixels, MGA had 97.89% recognition accuracy with training time of 61.20ms while MCPN gave 97.44% recognition accuracy in a time of 62.46ms achieved. At 2480, MGA had 96.67% with a training time of 4.53ms, whereas MCPN had 96.33% accuracy with a time of 4.98ms achieved. Furthermore, at 1240 database sizes, MGA has 96.44 % recognition accuracy with 0.62ms training time whereas MCPN gave 96.11% accuracy with 0.75ms training time. The two techniques were evaluated using different performance metrics. The results suggested the superior nature of the MGA technique in terms of epoch, recognition accuracy, convergence time, training time and sensitivity.
... A number of methodologies have been proposed over the years for online character recognition and some of them have outstanding performance. [11] developed a genetic based neural network model for online character recognition. This research integrated modified genetic algorithm into modified backpropagation neural network to improve the performance of online character recognition system. ...
Online character recognition is characterized with feature extraction and classification parameters
that make recognition accuracy non-trivial task. Failure of existing optimization techniques to yield
an acceptable solution to solve poor feature selection and slow convergence time provokes the
idea for some stochastic algorithms. In this paper, a feature reduction technique that apply the
power of genetic algorithm was modified using fitness function and genetic operators to minimize
the aforementioned drawbacks. Two classifiers (C1 and C2) were then formulated from the
integration of modified genetic algorithm (MGA) into an existing Modified Optical Backpropagation
(MOBP) learning algorithm. The performance of C2 on generation gaps was further evaluated
using convergence time and recognition accuracy. The research evaluation showed that C2
assumed average convergence times of 130.30, 211.69, 199.23 and 243.00 milliseconds with
generation gaps of 0.1, 0.3, 0.5 and 0.7. This implies that generation gap variation had a positive
effect on the network performance. Further evaluation showed that C2 assumed average
recognition accuracies at 0.7 is 98.1% and 99.4% at Ggap 0.1 respectively.
... Character recognition is an ongoing research which has motivated researchers from various aspect of human endeavors such as image processing, computer vision and machine learning [1]. Online character recognition system is the transformation process, that will allow extraction of input characters, from character image database to digitalize and translates the handwritten text into a machine editable form [2]. Due to global security threat, person authentication, and retrieving text, there is need of adopting techniques that could enhance the recognition performance of the system. ...
... In this article, the first step is the character images were acquired using a pen digitizer from datasets of 6200 samples collected by (Adigun et al., 2016) was used for training. Three preprocessing techniques were employed: binarization, extreme coordinate measurement and grid resizing were used to convert into binary form, measure extreme coordinate of the space and matrix standard respectively. ...
This paper carries out performance evaluation of a Modified Genetic Algorithm (MGA) and Modified Counter Propagation Network (MCPN) Network for online character recognition. Two techniques were used to train the feature vectors using supervised and unsupervised methods. First, the modified genetic algorithm used feature selection to filter irrelevant features vectors and improve character recognition because of its stochastic nature. Second, MCPN has its capability to extract statistical properties of the input data. MGA and MCPN techniques were used as classifiers for online character recognition then evaluated using 6200-character images for training and testing with best selected similarity threshold value. The experimental results of evaluation showed that, at 5 x 7 pixels, MGA had 97.89% recognition accuracy with training time of 61.20ms while MCPN gave 97.44% recognition accuracy in a time of 62.46ms achieved. At 2480, MGA had 96.67% with a training time of 4.53ms, whereas MCPN had 96.33% accuracy with a time of 4.98ms achieved. Furthermore, at 1240 database sizes, MGA has 96.44 % recognition accuracy with 0.62ms training time whereas MCPN gave 96.11% accuracy with 0.75ms training time. The two techniques were evaluated using different performance metrics. The results suggested the superior nature of the MGA technique in terms of epoch, recognition accuracy, convergence time, training time and sensitivity.
... OCR (offline) recognizes handwritten text that have been previously typed or scanned prior to recognition process or once the writing process is over [10,11] whereas in ICR (online), handwritten data are captured and recognized when character are under creation or during the writing. ICR is superior over OCR due to the temporal information available with the online, the two dimensional coordinates of successive points are represented as function of time and information on the ordering of strokes are available [12,13,14]. Many classifiers were used for character recognition, but Artificial Neural Networks (ANNs) outperform the other classifiers because of its flexibility, scalability, accuracy and learning. ...
_______________________________________________________________________________ Abstract Feature extraction and feature selection place an important role in online character recognition and as procedure in choosing the relevant feature that yields minimum classification error. Character recognition has been a good research area for many years because of its potential applications in all the fields. However, most existing classifiers used in recognizing online handwritten characters suffer from poor selection of features and slow convergence which affect recognition accuracy. A genetic algorithm was modified through its fitness function and genetic operators to minimize the character recognition errors. In this paper Modified Genetic Algorithm (MGA) was used to select optimized feature subset of the character to extract discriminant features for classification task. Some of research works have tried to improve online character recognition and their works were based on learning rate and error adjustment which slow down the training process. Thus, to alleviate this problems, a genetic based neural network model was developed using MGA to optimize an existing Modified Optical Backpropagation (MOBP) neural network. Two classifiers (C1 and C2) were formulated from MGA-MOBP such that C1 classified using MGA at classification level while C2 employed MGA at both feature selection level and classification level. The experiment results showed that the developed C2 achieved a better performance with no recognition failure and 99.44 recognition accuracy
... In this paper MGA was used to find the optimal feature subset. The reasons for performing feature selection as stated by [8,9] was to reduce computational time, improving data understanding and better classification performance. The generation gap is the proportion of chromosomes in the population which are replaced in each generation. ...
Online character recognition is characterized with feature extraction and classification parameters that make recognition accuracy non-trivial task. Failure of existing optimization techniques to yield an acceptable solution to solve poor feature selection and slow convergence time provokes the idea for some stochastic algorithms. In this paper, a feature reduction technique that apply the power of genetic algorithm was modified using fitness function and genetic operators to minimize the aforementioned drawbacks. Two classifiers (C1 and C2) were then formulated from the integration of modified genetic algorithm (MGA) into an existing Modified Optical Backpropagation (MOBP) learning algorithm. The performance of C2 on generation gaps was further evaluated using convergence time and recognition accuracy. The research evaluation showed that C2 assumed average convergence times of 130.30, 211.69, 199.23 and 243.00 milliseconds with generation gaps of 0.1, 0.3, 0.5 and 0.7. This implies that generation gap variation had a positive effect on the network performance. Further evaluation showed that C2 assumed average recognition accuracies at 0.7 is 98.1% and 99.4% at Ggap 0.1 respectively. Short Research Article Adigun et al.; BJAST, 18(5): 1-8, 2016; Article no.BJAST.31277 2
