ArticlePDF Available

1x1 Trainer with Handwriting Recognition


Abstract and Figures

Nowadays, computers and mobile devices play a huge role in our daily routines; they are used at work, for private purposes and even at school. Moreover, they are used as support for different kinds of activities and task, like for example, learning applications. The interaction of these applications with a computer is based on predefined input methods, whereas a touchscreen facilitates direct input via handwriting by using a finger or a pen. This paper deals with the invention of a mobile learning application, which is supposed to facilitate children’s learning of simple multiplication. The aim of this paper is to collect the data of children’ experiences using interactive handwriting on mobile devices. In order to gain this data, a school class of the school “Graz-Hirten” was tested and afterwards for evaluational purposes interviewed. The results of these usability tests have shown that children perceived handwriting via finger on screen as quite positive.
Content may be subject to copyright.
Paper1x1 Trainer with Handwriting Recognition
1x1 Trainer with Handwriting Recognition
M. Rabko, and M. Ebner!!"
Graz University of Technology, Graz, Austria
AbstractNowadays, computers and mobile devices play a huge role in our
daily routines; they are used at work, for private purposes and even at school.
Moreover, they are used as support for different kinds of activities and task, like
for example, learning applications. The interaction of these applications with a
computer is based on predefined input methods, whereas a touchscreen facili-
tates direct input via handwriting by using a finger or a pen.
This paper deals with the invention of a mobile learning application, which
is supposed to facilitate children’s learning of simple multiplication. The aim of
this paper is to collect the data of children’ experiences using interactive hand-
writing on mobile devices. In order to gain this data, a school class of the school
“Graz-Hirten” was tested and afterwards for evaluation purposes interviewed.
The results of these usability tests have shown that children perceived handwrit-
ing via finger on screen as quite positive.
Keywordshandwriting recognition, convolutional neural network, multiplica-
tion table, learning, children, mobile learning
1 Introduction
Handwriting is an individual skill and used for communication such as sharing
emotions and opinions. Everybody has their own individual way and style of writing
letters, numbers and signs. Despite the fact that every person writes quite individually,
people recognize and understand what is written. Keypads make it possible to enter
letters, numbers and signs into electronic devices. Because of predefined and pre-
programmed input methods, these devices are able to process written language. How-
ever, input via handwriting makes this process quite more difficult since devices can-
not process or interpret it.
Thus handwriting recognition is a subarea of pattern recognition. The idea is to
“teach” a device how to process and recognize handwriting. Devices are supposed to
notice changes in their environment and to recognize patterns and subsequently to
make decisions [1].
In the recognition of handwriting it can be differentiated between “online” and “of-
fline” recognition [2]. On the one hand, in relation to online recognition, handwriting
is being processed during the input. The information that is collected during this pro-
iJIM Vol. 12, No. 2, 2018
Paper1x1 Trainer with Handwriting Recognition
cess is of great importance. On the other hand, in relation to offline recognition, the
information is only processed and recognized at a later point.
Handwriting recognition is used in many different areas such as Personal Digital
Assistants (PDA), banking, post services [3] and even learning applications for chil-
The aim of this paper is to show how young schoolchildren react to this input
method and to evaluate test results of usability tests of handwriting recognition.
Moreover, it will be examined if this input method is favoured by children when it is
compared to the conventional keypad input method. In order to get these results, a
learning application with included offline handwriting detection has been created and
evaluated. The invented application should facilitate children learning of simple mul-
tiplication and is based on the already existing application “1x1 trainer” which was
created by Graz University of Technology. The app itself will be another Learning-
Analytics-Application [14] because each calculation will also be sent to a central
database and analysed for further trainings.
2 Prototyping
From a research perspective the study is following the method of prototyping with
proof of concept afterwards. With other words an application was built which fulfils
the defined goals and was given to children for a first field test. After a thoroughly
study of the app store only few math apps were found with the possibility of hand-
writing numbers, namely: ABCFunkid Calculus Lite, Write Math, Handwriting Math
Training, Handwriting Math Training for Kids and Todo Math. All apps in common
were that the handwriting works not perfect and so an improvement seems to be obvi-
The already mentioned „1x1 trainer“ is a mathematical learning application which
facilitates the learning and studying of simple multiplications [4]. The special feature
of the new developed app is that it is able to recognize and process handwritten num-
bers. Answers to the calculation tasks can be written onto the touchscreen of a mobile
device. The application is created for a target group of children between 6 and 12
years. One main idea was that children do not get distracted or confused by too much
text [5]. Children who use this app should be attracted and motivated by the design of
the user interface. Therefore, the main focus of this application was the clear struc-
tured user interface and the abandonment of visual elements.
The application has two playing modes, which share the same principle. The only
difference between these two modes is the variety of tasks and the option of saving
the learning progress. In the first mode, a server, which runs in the background,
chooses the tasks with the help of an algorith. In contrast, in the other mode tasks are
chosen randomly.
There are 20 calculation tasks, which the player is supposed to answer within 60
seconds. The faster the task is solved, the more points can be scored. At the end of the
game, the solved tasks and the gained points are summed up which is supposed to
support the learning effect.
Paper1x1 Trainer with Handwriting Recognition
3 Application
The newly invented learning game uses for the recognition of handwritten numbers
a Convolutional Neural Network consisting of multiple layers since this kind of neu-
ral network scores the best results concerning the recognition of handwritten numbers
[7][8]. The so-called CNN consists out of two Convolution-layers, to which the Max-
Pooling-layer is connected, and in addition, a fully connected layer. The Figure 1
shows the structure of the used CNN.
Fig. 1. Convolutional neural network
3.1 Training
The defined CNN has to be trained with suitable and sufficient data in order to rec-
ognise handwritten numbers. TensorFlow and the MNIST database are used for this
process of training the neural network. TensorFlow is a machine-learning library
developed by Google [9]. This library builds the neural network and provides it with
input of the MNIST database. The MNIST database consists of images of handwritten
numbers of the size of 28x28 pixel. It includes 60.000 images for the training and
further 10.000 images to test the new-configured neural network [10].
The duration of the whole process described here depends on the processing power
of the device and takes approximately 30 minutes. The CNN scores a recognition rate
of 99% over all test images.
3.2 Adaptation
The TensorFlow script saves weights and bias of all layers of the trained convolu-
tional neural network in separate files. Therefore, in order to apply the CNN with the
saved files, the 1x1 trainer” uses the “Basic neural network subroutines” (BNNS).
Apple introduced the BNNS with the iOS 10. It is part of the Accelerate framework
and helps the developer to run neural networks in iOS-applications. However, BNNS
supports only neural networks, which are already trained with test data.
iJIM Vol. 12, No. 2, 2018
Paper1x1 Trainer with Handwriting Recognition
The import of the data written with TensorFlow carries the risk that the neural net-
work does not function properly. There are differences between TensorFlow and
BNNS concerning the arrangement of pixel values within an array. BNNS expects
pixel values to be arranged in row by row, by channels within a file. TensorFlow
arranges data in a completely different way images are arranged in pixel per channel
by pixel per channel. In Figure 2 the differences in the arrangement of image data are
shown. Applied matrix operations within the training script solve these problems.
The writing of numbers is allowed in two predefined boxes whereas each box rep-
resents a number. The advantage of this is that the application does not have to filter
single numbers from the input.
Moreover, the application is, due to this, clearly arranged and easy to operate. The
Figure 3 shows these two boxes for the input of handwritten numbers.
Fig. 2. Difference of the data representation between BNNS and Tensorflow.
Fig. 3. 1x1 Trainer Trainer mode
Paper1x1 Trainer with Handwriting Recognition
After the user has written/drawn into one of the boxes, an UIImage is being pro-
cessed. In the following step, the UIImage gets preprocessed for the recognition. This
preprocessing includes the centering of the written number in the image and subse-
quently the scaling of the UIImage. This scaling is necessary because the convolu-
tional neural network accepts only images sized 28x28 pixel.
After the preprocessing, the recognition starts with the UIImage. In the next step,
the image runs through the following steps:
1. The first layer is a convolutional layer and calculates 32 features for each of the
5x5 kernel.
2. The second layer is a max-pooling layer with a 2x2 filter. This layer scales the im-
age to 14x14 pixel.
3. In the next step, another convolutional-layer calculates 64 features for each of the
5x5 kernel.
4. Afterwards, another max-pooling layer with a 2x2 filter scales the image to 7x7
5. After these four layers, a fully connected layer can be found. This layer helps with
the processing of the whole image.
6. In the last step, a softmax layer produces out of the resulting vector of the previous
layer a vector with 10 entries. These ten entries represent numbers from zero to
3.3 General architecture of the app
The application itself consist of 6 main screens, where users can navigate by press-
ing a back or forward button
1. Start screen: The user is welcomed and can choose between the trainer mode
(logged-in mode) or just the training mode
2. Login screen: If the user chose the trainer mode he/she can provide his/her creden-
tials. Afterwards each singe calculations will be sent to the server and saved for
further learning analytics operations [13]
3. Onboarding screen: This screen is just for testing handwriting for the very first
time. With other words a kind of help screen to show how to write the numbers of
the result on the right place.
4. Main screen: The main screen (see Fig. 3) consist of the provided calculation and
two boxes, where the result should be written by hand. Furthermore a small clock
is presented as well as the current level points. Two buttons allow the use to con-
firm his/her result or to delete it (for example if the numbers were wrongly inter-
preted by the algorithm)
5. Summary screen: This screen summarizes the result of the trial how many calcu-
lations done and how many of them correct. Finally the given points are displayed
6. High-score screen: In the end a high score list is presented
iJIM Vol. 12, No. 2, 2018
Paper1x1 Trainer with Handwriting Recognition
4 Evaluation
The evaluation was conducted with an iPad-class of the elementary school Graz-
Hirten following consequently a multi-level evaluation using the cut-off technique
[11] [12]:
First of all children are monitored by a research assistant during using the applica-
tion in the classroom
Second, interviews with the children were carried out in groups of 3-4 using the
cut-off technique, due to children often are shy and did not express their real feel-
ings about the application. Therefore a couple of statements were provided (see
Table 1) and the pupils had to rank those statements using different symbols (smi-
leys) providing their opinions against each other. So the research assistant did not
directly ask his questions, but take the discussion amongst them as basis for his
evaluation of the app.
For the final evaluation unfortunately only seven iPads could be used for testing
the application since the other four iPads did not have the required iOS-version 10
installed. On the day where the evaluation took place, 20 of 25 pupils were attendant.
The pupils were divided into three groups of 6-7. Each group played for 15 minutes
with the 1x1 trainer”. However, the game had not been explained to the pupils be-
forehand. The reason for this was that the pupils should not be influenced by anything
and come to their own opinion on this game since the latter is crucial for the opinion
Fig. 4. Evaluation setup
Paper1x1 Trainer with Handwriting Recognition
After each group had the chance to play with the “1x1 trainer”, the pupils were di-
vided up again into smaller groups of 3-4. At the end, six groups of pupils were asked
to provide feedback on the game. This feedback makes also the main part of the eval-
uation of the learning application. The pupils were asked to pick one of five smileys
(as it can be seen in the Table 1) in order to assess a statement to be true for them-
selves (laughing smiley) or not true for themselves (sad smiley). The statements were
carefully selected based on our experiences in similar research studies [11] [12].
These following statements had to be assessed by the pupils:
I liked the game.
I was able to get along with the game.
Drawing numbers with my finger is fun for me.
I would like to play the 1x1 trainer at home as well.
Table 1. Rating smileys
very good (1)
good (2)
normal (3)
bad (4)
very bad (5)
Table 2 gives an overview about the answer of the six groups concerning the given
statements. The result of each group is pointed out as well as the average over all
Table 2. Evaluation results for each group
Group 2
Group 5
Group 6
I liked the game
I was able to get along with
the game 2 3 1 2 1 2 1,83
Drawing numbers with my
finger is fun for me
1 1 1 3 2 1 1,5
I would like to play the 1x1
trainer at home as well
1 1 3 4 1 1 1,83
The research assistant also collected some statements of the children (Fig. 5), like
“the game was fun” and “very interesting to play during a math lesson”. Concerning
the handwriting he recognized that “it is fun to write numbers by hand” but also “that
iJIM Vol. 12, No. 2, 2018
Paper1x1 Trainer with Handwriting Recognition
it become stressful if the number is not written exactly”. Some children can imagine
“to play the game also at home”, but “maybe it can become boring too”. Furthermore
children summarized that the “app is great for an iPad, but handwriting on iPhone is
cumbersome”. Overall it can be stated that they liked th app, also the idea of hand-
writing and just mentioned some concerns about the small display of an iPhone and
few problems not-recognized numbers.
Fig. 5. Evaluation result of one group
5 Discussion
All in all, the evaluation of the prototype could be conducted without any greater
problems. The “1x1 trainer” as well as the connection to the server was functioning
properly. The pupils had a lot of fun drawing numbers with their fingers. Even though
the learning application got mainly positive feedback, the evaluation has also shown
that the usability of the application needs improvement since several problems oc-
curred, such as:
Recognition of handwriting. Sometimes, written numbers were recognized
wrongly which led to some confusion and frustration of the pupils. Especially the
numbers 1 and 7 were confused by the application.
Login in trainer mode. The login was a problem for some of the children since
they tried to log in with their true names even though they had got the user data
shortly before the test.
Playing the game. Some pupils were overstrained when they started the game and
did not know what to do with it.
Single-digits. Another problem, which confused the children, was the question in
which of the two boxes one-digit solutions should be written. However, the 1x1
traineraccepts both boxes.
Paper1x1 Trainer with Handwriting Recognition
6 Conclusion
The outcome of the evaluation shows clearly the potential of applications with
handwriting recognition; especially of learning applications, which dispose this func-
tion. The pupils had a lot of fun drawing the numbers onto the screen. However, their
opinions on whether they preferred handwriting or writing on a keypad were diverse.
A reason why some of the pupils did not like handwriting could be the problems they
had with the handwriting recognition since wrongly read numbers had a vast influ-
ence on the pupils’ motivation.
Moreover, a positive side effect of using handwriting recognition in this applica-
tion was that children put in more effort into drawing the numbers readable. This side
effect was found due to the feedback survey and the assessment given by pupils.
In conclusion, the research for this paper supports that handwriting recognition
should be used in learning applications for children, because it seems to motivate
them. However, the success of an application strongly depends on the recognition rate
as well as on the applied methods.
Of course the limitation of this study is that we just examined a prototype and its
usability issues. The study did not examine if handwriting of numbers leads to a deep-
er learning or not. So we concentrate just on making an application more fun and easy
to use. In further studies it must be investigated if those apps and learning analytics
analyses help to learn the multiplication table more efficiently as first attempts have
already shown [15].
Because of the mentioned problems with the “1x1 trainer”, some features of the
application should be improved. The following approaches provide some ideas for
improvements and extensions:
Pretest. The1x1 Trainerdoes not use a pretest on the trainer mode yet to evalu-
ate the existing knowledge of the user. The result of the lack of a pretest is that
every user has to start with the simplest tasks. A pretest would motivate older pu-
pils and those who are already quite good at simple multiplication.
Handwriting Recognition. Although, the handwriting recognition is working with
the application, there is still improvement possible. It might help to configure the
applied neural network better or to extend it. Another possibility to improve the
recognition would be to apply additional pre-processing techniques.
Explanation of the Game. The playing modes are only explained in the main
menu while the exact distribution is never mentioned at any point.
Gamification. In order to maintain the motivation of the user, it would help to use
gamification elements. Numbers that were read wrongly would not have such a
huge impact on the playersmotivation.
iJIM Vol. 12, No. 2, 2018
Paper1x1 Trainer with Handwriting Recognition
7 References
[1] A. K. Jain, R. P. W. Duin, and J. Mao, „Statistical pattern recognition: A review,“ IEEE
Transactions on pattern analysis and machine intelligence, vol. 22, no. 1, pp. 4–37, 2000.
[2] R. Plamondon and S. N. Srihari, „Online and off-line handwriting recognition: A compre-
hensive survey,“ IEEE Transactions on pattern analysis and machine intelligence, vol. 22,
no. 1, pp. 63–84, 2000.
[3] A. Priya, S. Mishra, S. Raj, S. Mandal, and S. Datta, „Online and offline character recogni-
tion: A survey,“ in Communication and Signal Processing (ICCSP), 2016 International
Conference on, IEEE, 2016, pp. 0967–0970.
[4] Schön, M., Ebner, M., Kothmeier, G. (2012) It's Just About Learning the Multiplication
Table, In Proceedings of the 2nd International Conference on Learning Analytics and
Knowledge (LAK '12), Simon Buckingham Shum, Dragan Gasevic, and Rebecca Ferguson
(Eds.). ACM, New York, NY, USA, 73-81.
[5] Huber, S., & Ebner, M. (2013). iPad Human Interface Guidelines for M-Learning. In Z.L.
Berge and L.Y. Muilenburg (Eds.), Handbook of mobile learning. (pp. 318-328). New
York: Routledge
[6] Ebner, M. (2015) Mobile Learning and Mathematics. Foundations, Design, and Case Stud-
ies. Crompton, H., Traxler, J. (ed.). Routledge. New York and London. pp. 20-32
[7] P. Y. Simard, D. Steinkraus, J. C. Platt, et al., „Best practices for convolutional neural
networks applied to visual document analysis.,“ in ICDAR, vol. 3, 2003, pp. 958–962.
[8] D. Ciregan, U. Meier, and J. Schmidhuber, „Multi-column deep neural networks for image
classification,“ in Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Confer-
ence on, IEEE, 2012, pp. 3642–3649.
[9] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis,
J. Dean, M. Devin, et al., „Tensorflow: Large-scale machine learning on heterogeneous
distributed systems,“ arXiv preprint arXiv:1603.04467, 2016.
[10] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, „Gradient-based learning applied to doc-
ument recognition,“ Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998.
[11] Ebner, M., Schönhart, J. & Schön, S. (2014). Experiences with iPads in primary schools.
In: Profesorado, V 18, Nº 3, 161-173 (sept.-diciembre 2014).
[12] Lexow, S., Ebner, M. (2014) Development of a Collaborative Learning Game Using Ex-
ternal Plastic Cards as an Input Device on an iPad, International Journal of Interactive Mo-
bile Technologies (iJIM), Vol. 8, Nr. 2, p. 12-17
[13] Kraja, E., Taraghi, B. & Ebner, M. (2017). The Multiplication Table as an innovative
Learning Analytics Application. In J. Johnston (Ed.), Proceedings of EdMedia: World
Conference on Educational Media and Technology 2017 (pp. 810-820). Association for the
Advancement of Computing in Education (AACE).
[14] Ebner, M. (2015) Mobile Learning and Mathematics. Foundations, Design, and Case Stud-
ies. Crompton, H., Traxler, J. (ed.). Routledge. New York and London. pp. 20-32
[15] Ebner, M., Schön, M., Neuhold, B. (2014). Learning Analytics in basic math education
first results from the field, eLearning Papers, 36, pp. 24-27
Paper1x1 Trainer with Handwriting Recognition
8 Authors
M. Rabko studied Software Engineering and Management at Graz University of
Technology (e-mail:
M. Ebner is the head of Department Educational Technology at Graz University
of Technology and therefore responsible for all university wide e-learning activities.
He holds an Adjunct Prof. on media informatics (research area: educational technolo-
gy) and works also at the Institute of Interactive Systems and Data Science as senior
researcher. His research focuses strongly on seamless learning, learning analytics,
open educational resources, making and computer science for children. In 2012 he
was awarded the New German Book Prize. 2015 he received the Austrian State Prize
for Adult Education for his online course “Gratis Online Lernen” on the “iMoox” e-
learning platform. For publications as well as further research activities, please visit
his website:
Article submitted 18 September 2017. Resubmitted 25 December 2017 and 10 January 2018. Final ac-
ceptance 06 March 2018. Final version published as submitted by the authors.
iJIM Vol. 12, No. 2, 2018
... Several researches have already tried to embed computer vision technology into smartphone and proved that it can be a more convenient alternative to a computer-based system. For example, mobile learning application to facilitate children in learning of simple multiplication, Javanese letters classifier with mobile client-server architecture, and smartphone-based image color correction for color blindness [12][13] [14]. ...
Full-text available
This research focuses on developing an Android-based cattle identification system that is applicable and easy to use. This system uses a Scale Invariant Feature Transform (SIFT) algorithm to extract features from the muzzle images, and Random Sample Consensus (RANSAC) algorithm to eliminate features incompatibility. The system is experimented with four threshold values, i.e. 10, 15, 20, and 30 using a total data of 460 muzzle images. In the first experiment, 3 images from each individual are used in the training stage and 2 images are used as the data test. In the second experiment, 5 images from each cattle are used in the training stage and 5 images are used as the data test. Data used in training stage are 244 images and in testing stage is 816 images. From the experiment, the highest accuracy rate is 98.1% with threshold values of 15 and 20. The execution time is also calculated to measure the processing time of the system. The average time taken to store an image to the database is 1.3 seconds. The main contribution of this research is technology implementation and more portable muzzle identification for local cattle in Makassar.
Conference Paper
Full-text available
The main topic of this paper is the development of a web-based application that helps children to learn the one-digit multiplication table. The developed application supports individual learning process of the pupils and also provides the teachers with the possibility to intervene according to the analysis of users’ answers. The application uses modern technologies in order to offer high performance and availability to the users. The system also provides an interface for mobile clients, which present the questions and the processed data in different forms. The answers of the pupils, as well as other gathered data from the application show interesting results related to the participation and learning improvement.
Full-text available
TensorFlow is an interface for expressing machine learning algorithms, and an implementation for executing such algorithms. A computation expressed using TensorFlow can be executed with little or no change on a wide variety of heterogeneous systems, ranging from mobile devices such as phones and tablets up to large-scale distributed systems of hundreds of machines and thousands of computational devices such as GPU cards. The system is flexible and can be used to express a wide variety of algorithms, including training and inference algorithms for deep neural network models, and it has been used for conducting research and for deploying machine learning systems into production across more than a dozen areas of computer science and other fields, including speech recognition, computer vision, robotics, information retrieval, natural language processing, geographic information extraction, and computational drug discovery. This paper describes the TensorFlow interface and an implementation of that interface that we have built at Google. The TensorFlow API and a reference implementation were released as an open-source package under the Apache 2.0 license in November, 2015 and are available at
Full-text available
Tablet computers gain enormous attention nowadays and become more and more part of our daily life. Due to this it is not astonishing that even the educational sector is thinking about the use of such wearable devices in the classroom. Our research study aims to give insights about real life experiences with iPads in Austrian primary schools. Therefore we describe the development of appropriate learning apps and their use in classrooms. Finally each technology-enhanced lesson is observed as well as evaluated afterwards with the help of the cut-off technique. The research work carries out different circumstances the use of tablet computers has a positive influence on teaching and learning and gives practical hints how they should be used in classrooms.
Full-text available
This paper evaluates and describes the usage of plastic cards, coated with conductive paint, as an input device for capacitive touchscreens. By using the developed card prototypes it should be proofed, that usage of this new kind of input device can be handled by primary school pu- pils in a collaborative learning game. For this reason an educative digital learning game has been developed which can be controlled by the card prototypes. The game asks questions of general knowledge and the answer can be given by putting the proper plastic card on the touchscreen. The evaluation of the game by two groups of four children pointed out, that the cards can be easily used to identify a specific user. Although evaluation shows that the card con- trol has weaknesses to reliably detect the correct answer during the game phase. All pupils enjoy to play the game and they additionally state, that they like the usage of the cards. Beside the problems with a reliable card recognition the evaluation shows that the collaborative concept of the game is promising due to the fact that the pupils are always work- ing together on finding a solution for the answer. Further they support each other in handling and understanding the plastic cards which leads to a deeper understanding of the technical backgrounds.
Full-text available
Learning Analytics is an emerging field, because the analyses of a big amount of data can lead to deeper insights into how learning occurs. In this publication we introduce a new addition / subtraction trainer which assists teachers in their daily basic math education. A first field study points out that beside of typical mistakes also systematic ones can be detected.
Conference Paper
Full-text available
One of the first and basic mathematical knowledge of school children is the multiplication table. At the age of 8 to 10 each child has to learn by training step by step, or more scientifically, by using a behavioristic learning concept. Due to this fact it can be mentioned that we know very well about the pedagogical approach, but on the other side there is rather less knowledge about the increase of step-by-step knowledge of the school children. In this publication we present some data documenting the fluctuation in the process of acquiring the multiplication tables. We report the development of an algorithm which is able to adapt the given tasks out of a given pool to unknown pupils. For this purpose a web-based application for learning the multiplication table was developed and then tested by children. Afterwards so-called learning curves of each child were drawn and analyzed by the research team as well as teachers carrying out interesting outcomes. Learning itself is maybe not as predictable as we know from pedagogical experiences, it is a very individualized process of the learners themselves. It can be summarized that the algorithm itself as well as the learning curves are very useful for studying the learning success. Therefore it can be concluded that learning analytics will become an important step for teachers and learners of tomorrow.
Full-text available
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
Full-text available
The primary goal of pattern recognition is supervised or unsupervised classification. Among the various frameworks in which pattern recognition has been traditionally formulated, the statistical approach has been most intensively studied and used in practice. More recently, neural network techniques and methods imported from statistical learning theory have been receiving increasing attention. The design of a recognition system requires careful attention to the following issues: definition of pattern classes, sensing environment, pattern representation, feature extraction and selection, cluster analysis, classifier design and learning, selection of training and test samples, and performance evaluation. In spite of almost 50 years of research and development in this field, the general problem of recognizing complex patterns with arbitrary orientation, location, and scale remains unsolved. New and emerging applications, such as data mining, web searching, retrieval of multimedia data, face recognition, and cursive handwriting recognition, require robust and efficient pattern recognition techniques. The objective of this review paper is to summarize and compare some of the well-known methods used in various stages of a pattern recognition system and identify research topics and applications which are at the forefront of this exciting and challenging field
Full-text available
Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day
Conference Paper
Handwritten character recognition has been one of the most fascinating research among the various researches in field of image processing. In Handwritten character recognition method the input is scanned from images, documents and real time devices like tablets, tabloids, digitizers etc. which are then interpreted into digital text. There are basically two approaches — Online Handwritten recognition which takes the input at run time and Offline Handwritten Recognition which works on scanned images. In this paper we have discussed the architecture, the steps involved, and the various proposed methodologies of offline and online character recognition along with their comparison and few applications.