Neural Networks for Handwriting Recognition

Contents 

1 Executive Summary ................................................................................................................. 

2 Introduction/Background ........................................................................................................ 

3 Main Body/Content ............................................................................................................... 

3.1 Historical Survey .............................................................................................................

3.2 State of the art ............................................................................................................... 

3.3 Case Studies ................................................................................................................... 

3.3.1 Handwriting Recognition using Neural Network .................................................... 

3.3.2 Handwritten Character Recognition using Neural Networks ................................. 

3.3.3 Neural Networks for Handwritten English Alphabet Recognition ..........................

3.3.4 Diagonal Based Feature Extraction for Handwritten Alphabets Recognition System using Neural Network ............................................................................................................ 

3.3.5 Image preprocessing for optical character recognition using neural networks ..... 

3.3.6 Intelligent Systems for Off-line Handwritten Character Recognition: A Review... 

3.3.7 Fuzzy Based Handwritten Character Recognition System ...................................... 

3.3.8 Recognition for Handwritten English Letters: A Review ......................................... 

3.3.9 An Overview of Character Recognition Focused on Off-Line Handwriting ............ 

3.3.10 Offline Handwritten English Numerals Recognition using Correlation Method... 

3.4 Future Prospects ............................................................................................................ 

4 Conclusions ............................................................................................................................ 

5 References .............................................................................................................................

 

 

 

 

 

  

 

1 Executive Summary 

Human's intelligence makes them separate from computers. The different chores that are yet not possible for the machine to do it on its own; but are been able to be done by a human. One of the chores is handwriting recognition. Although, handwriting recognition in the handwritten documents has been considered as one of the important areas of the research over the past few decades and due to which numerous automatic handwritten systems are evolved by various researchers in earlier times. The handwriting recognition system generally composed of many phases amongst which preprocessing, feature extraction, and classification are the main steps in the available approaches. Artificial Neural Network can be demonstrated as the rescuer in the evolvement of an accurate and efficient handwritten text recognition system. Maybe pattern recognition is the most standard application of neural networks.[2] 

A neural network is a network or circuit of neurons, or in a modern sense, and artificial neural network, made up of nodes or artificial neurons.[3] Handwriting recognition is defined as the potential of a machine to accept and simplify handwritten input from various sources like documents, images, smartphones, etc.[4] Conventionally handwriting recognition is divided into online and offline recognition.[5] The automatic transformation of letter codes from the text in a picture that is consumed in the applications that process text and inside the computer is involved in the off-line handwriting recognition. The automatic transformation of the text which is inscribed on Personal Digital Assistant or a special digitizer where a sensor detects pen-up/pen-down shifting and also the movements of the pen-tip is involved in the on-line handwriting recognition.[4] 

Conventional methods emphasize on dividing single characters for recognition, whereas modern techniques emphasize on recognizing each character in a sectioned line of text. Specifically, they emphasize on the methods of machine learning that can learn visual attributes, evading the restricting feature engineering used before. State-of-the-art methods use convolutional networks for extracting visual attributes in place of various overlapping windows of a text line image that Recurrent Neural Network makes use of for producing character chances.[4] 

Due to the methods of feature extraction like character geometry and gradient technique are utilized in the method does not categorize different language characters, the method can be elongated for the classification of languages independently from the images of other languages with few changes. For attaining a greater accuracy rate, the filtration of divided characters can be done. By including few more features besides the prevailing ones, the neural network performance can be increased. By training the neural network with a greater number of images for testing, the classification ratio can be enlarged.[1] The handwriting recognition using a neural network's work can be taken forward for other language character recognition.  Text format can be obtained by using it for the conversion of the fax and the newspapers. For classifying, multiple Artificial Neural Network can be used for recognizing words, sentences, or paragraphs. Also, for reading postal addresses in the post office it can be used.[6] 

 

Handwriting systems can be used for recognizing plenty of different regional languages having distinct writing styles applying respective algorithms and strategies. Because of the strange character's presence or resemblance in shapes of various characters, it has emerged that handwritten character recognition becomes difficult. The characters are divided into single characters and for getting a clean image, the scanned image is pre-processed. Normalization, filtration is done with the usage of processing steps results in clean and noise-free outcome under pre-processing work. The system desired result with effectiveness can be achieved by governing algorithms with assessment, proper training, and other step-by-step procedure. For achieving finer English characters recognition output, some geometric and statistical features must be used.[6] Artificial Neural networks can be used in handwriting recognition which can be done using a feed-forward backpropagation algorithm that is applied to the collection of data at a specific time. Handwriting recognition can be done effectively by using proper methods with accuracy and method complexity. Handwriting recognition can be done by using feature extraction, clustering, matching of the pattern; but it can be said as per the research done that the higher accuracy, efficiency, and reliability can be achieved by using neural networks.[5] 

 

2 Introduction/Background 

Human's intelligence makes them separate from computers. The different chores that are yet not possible for the machine to do it on its own; but are been able to be done by a human. One of the chores is handwriting recognition. Although, handwriting recognition in the handwritten documents has been considered as one of the important areas of the research over the past few decades and due to which numerous automatic handwritten systems are evolved by various researchers in earlier times. Though, the algorithm of the recognition and its effectiveness is yet an open research topic. Because of the extensive irregularity in handwriting types, the start-of-the-art handwriting recognition systems often fail in providing adequate execution on different handwriting examples types. The handwriting recognition system generally composed of many phases amongst which preprocessing, feature extraction, classification, and post-processing are the main steps in the available approaches. Feature extraction and classifier design are the two important phases of any recognition system, however. English, Chinese, Arabic, Japanese, Bangla, Malayalam, etc. are the several languages on which many researchers have made various types of handwritten text recognition systems. Yet the recognition issues of these languages’ scripts cannot be considered as totally resolved. Artificial Neural Network can be demonstrated as the rescuer in the evolvement of an accurate and efficient handwritten text recognition system. One of the prime ways through which computers are trained with human likeability is by using Artificial Neural Network in the design. Neural networks functions on the human brain design and they are specifically very helpful for resolving alike issues that cannot be expressed as a sequence of simple steps, like objects classification into several classes, recognition of patterns, prediction of series, and mining of data. Maybe pattern recognition is the most standard application of neural networks. The neural network is introduced with the corresponding input vectors that carry the pattern details as well as with a distinct class of intended vectors. The inputs can range from simple one-dimension (1-D) data to multidimensional data. The Artificial Neural Network can be utilized for discovering the class/patterns in the unseen data (latest inputs), as soon as it is trained under the train data like the human brain.[2] For research, recognition of handwritten and mechanism words is a growing area, and the usage in banks, offices, and industries is substantial. As compared to other architectures, design components are quite less identified due to neural computing is the latest technology analogously. The parallelism of data is applied by neural computers. The method in which neural computers are functioning is disparate as compared to normal computers functioning. A neural computer is not programmed but trained which gives a definite starting point (input of the data) that either makes the original data to develop which results in the optimization of the required property or arrange the data into a class number which has been inputted.[6] 

Neural Network: - A neural network is a network or circuit of neurons, or in a modern sense, and artificial neural network, made up of nodes or artificial neurons. For resolving artificial intelligence issues, a neural network is either an artificial neural network or a biological neural network composed of real biological neurons. The biological neuron connections are modeled as weights. A negative weight represents inhibitory connections, while the positive values indicate an excitatory connection. All the inputs are altered by mass and added. This process is mentioned as a linear combination. At last, an activation function manages the output amplitude. For instance, an allowable output range is generally in the middle of 0 and 1, or it could be -1 and 1. These artificial networks can be applied for the training via a dataset and may be used for adaptive control and predictive modeling. The experience can result in self-learning confines of networks, that can gain conclusions from a complex and information which is not related apparently.[3] 

 

The model that processes information which is influenced by the biological nervous systems like the brain processes information is called an artificial neural network. The novel structure of the information processing system is a vital factor in the model. For solving the issues, it is made up of a great number of highly processing interconnected elements (which is called a neuron) working together. Artificial Neural Network’s like learning, populating from instances. Through the process of learning, an artificial neural network is designed for a particular use like classifying data or recognition of the pattern. Biological system training includes adaptations to the neuromuscular junction which exists in between the neurons.[6] 

Feedforward neural network: - An artificial neural network in which connections between the nodes do not establish a series is defined as a feedforward neural network. Due to which feedforward neural network is different from its successor, which is recurrent neural networks. The feedforward neural network was the foremost and easiest type of artificial neural network conceived. As the name feedforward justifies, the information advances in a single direction only in the network that is from the input nodes across the hidden nodes if any available and finally to the output nodes. The loops or cycles are not present in the network.[7] It is one of the simplest types of artificial neural networks. The data moves from the various input nodes until it gets to the output nodes. This is generally accomplished by using a classifying activation function; therefore, it is also known as a front propagated wave. Backpropagation does not occur and thus data advances in a single direction only contrary to the other complex types of neural networks. A feedforward neural network may have concealed layers or may have a single layer. The total of the outcome of the input and their mass are calculated which is then given to the output in a feedforward neural network.[8] 

 

It is difficult to classify the desired classes in these applications due to which technologies such as computer vision and recognition of face making use of feedforward neural networks. The information that carries plenty of noise is dealt with by a simple feedforward neural network provided. Feedforward neural networks are also comparatively easy in maintenance.[8]  

Backpropagation neural network: - This backpropagation technique is utilized for training a multi-layer artificial neural network with the mathematical formulation. The main aim of this network is to train the network for balancing between the inputted patterns which are used in training and their input patterns responses. 

Backpropagation algorithm stages are: 

• Propagating the output solution back to the neural network 

• Updating the mass after every single propagation[5] 

Why Neural Network? 

Due to the extraordinary capability of extracting meaning from inaccurate or complex data, the neural network can be used for the deriving pattern and determining trend which is too complicated that cannot be observed by any computer approach or human. A neural network which has been trained can be considered as an “expert” in the classification of information which has been given for analyzing; which then can be used for providing estimate given new circumstances and answering the “what if” questions. 

 

 

Other benefits include 

• Adaptative Learning: The potential of learning how to work based on the given data for initial experience or training. 

• Self-Organization: Artificial Neural Network can create an organization on its own or the information representation which is received at the time of the learning. 

• Real-Time Operation: Artificial Neural Network computations may be implemented parallelly, and special hardware tools can be manufactured and designed which can take benefit of this potential. 

• Fault Tolerance Through Redundant Information Coding: Respective performance degradation can be attended via network partial demolition. Even with a crucial network deface some capabilities of the network may be maintained.[6] 

 

Handwriting recognition: - Handwriting recognition is defined as the potential of a machine to accept and simplify handwritten input from various sources like documents, images, smartphones, etc. Through intelligent word recognition or optical character recognition, the written text image may be sensed “offline”. As an alternative, the pen tip motions may be sensed “online”, for instance by a pen-based computer screen surface, normally an easy task due to the unavailability of more clues. A handwriting recognition executes accurate division into characters, controls formatting, and searches the most probable words. Online handwriting recognition can be fragmented into a few common steps like preprocessing, feature extraction, and classification.[4] 

Handwriting Recognition Implementation: - Handwritten Recognition functions using neural networks in steps like preprocessing, segmentation, feature extraction, and recognition. For making document images ready for the segmentation, preprocessing involves a sequence of functions to be done. The document image is segmented into the numeric image or individual character during segmentation then extraction of feature method is applied to it. For recognition, a feature vector is provided to the selected algorithm finally where the extracted features are presented to Neural Network for the recognition.[6] 

Off-line Recognition: -The automatic transformation of letter codes from the text in a picture that is consumed in the applications that process text and inside the computer is involved in the off-line handwriting recognition. The input acquired from this method is considered as a stable representation of handwriting. Due to various people having distinct handwriting styles, it is comparatively difficult for the off-line handwriting recognition and as of now, OCR(Optical Character Recognition) engines are mainly concentrated on the text that is machine printed and ICR (Intelligent Character Recognition) on the hand-printed text (text written in capital letters).[4] 

Online recognition: - The automatic transformation of the text which is inscribed on Personal Digital Assistant or a special digitizer where a sensor detects pen-up/pen-down shifting and also the movements of the pen-tip is involved in the on-line handwriting recognition. This type of data is recognized as digital ink and can be considered as a handwriting digital representation. The letter codes that are consumed in the computer and applications which processes text are transformed from the signal that is inputted. The components of an online handwriting recognition interface usually comprise: 

• A pen or style tool with which user writes 

• A touch-sensitive surface that may be merged with or adjoining an output display 

• A software application that explains the style tool’s motion on the surface of writing interprets into digital texts from the strokes’ outcome.[4] 

 

Advantages of Character Recognition: -  

• The proposal of Neural Network in Handwritten Recognition made us able to read varied merged style of a character writing. 

• Accuracy in character recognition with different size and font can be developed 

• Handwritten Character Recognition will be an effectual system for the gathering of the proof in the forensic application. 

• It also helps in reducing the noise from the original character. 

• Due to the testing session and heavy training, more sample sets lead to a greater rate of precision [6]. 

  

 

3 Main Body/Content 

Handwriting recognition is without any doubt one of the greatest difficult areas of the recognition of the pattern. To categorize alphanumeric or other optical types that are mainly stored as digital images is the objective of the optical character recognition. Based on the unique patterns, various pattern recognitions approaches have been used on both online and offline handwriting recognition. Various steps are involved in the recognition process. Neural network toolbox (special toolbox of MATLAB) that requires theoretical knowledge but reduces the difficulty of implementation is extensively used in areas like verification of the signature, processing of bank check, verification of document, etc. Optical Character Recognition is widely used in the entry of data. Due to the high forbearance of fault and parallel structure of Artificial Neural Network, its technique for the recognition of character is now acquiring significance.[5] 

In general, there are three general main steps involved in the recognition of handwriting: 

(1) Preprocessing which advances to segmentation, normalization, and noise-canceling 

(2) The image with the numerical feature vector is replaced with feature extraction for describing an image 

(3) By using the extracted feature, the classification stage with high precision tries recognition of handwritten character[9] 

The objective of preprocessing to get rid of the data which can harm the recognition and is not relevant to the data that has been inputted; which also concerns velocity and precision. Normalization, binarization, sampling, noise cancellation, and smoothing are generally included in the preprocessing. Feature extraction is the second step. Higher-dimensional data is extracted amongst the two or more-dimensional vector fields that are obtained from the preprocessing algorithms. Highlighting the prime information for the recognition model is the aim of the step. The data can include details like speed, the pressure of the pen, or the writing direction changes. Classification is the final important step. Different models are utilized for mapping the extracted to various classes and therefore recognizing the characters or the features represented words in this step.[4] 

3.1 Historical Survey 

Conventionally handwriting recognition is divided into online and offline recognition. A time prescribed succession of coordinates presenting the pen-tip motion is recorded in on-line recognition, whereas only a picture of the written work is obtained in an off-line instance. Online recognition normally obtains a good outcome due to the significant simplicity of taking out appropriate characteristics. The recognition of the text whole lines and that between the distinct characters or word recognition is the pivotal division. The latter is considerably tough as predicted and the magnificent outcomes which have been acquired for instance like digital and character recognition, have never coordinated for complete lines.  Finally, handwriting recognition can be divided into scenarios where the writing style is inhibited in some way. For instance, only hand-printed characters are allowed – and the more challenging case is where it is unrestricted. It remains an open issue producing an authentic, comprehensive system for unrestricted text line recognition despite research for more than 40 years of handwriting recognition.[10] 

Character extraction: - Scanning a form or document that has been written formerly is involved in the off-line character recognition which implies that the single characters will be needed to be taken out from the scanned image.  There are tools available that can perform this process. Although, there are various general defects in this step. Amongst which when characters are attached is considered as an individual sub-image comprising both characters, is a quite common mistake to have occurred during the procedure. That leads to an extensive issue in the recognition stage. Still, plenty of algorithms are there which can lessen the connected character's risk.[4] 

Character recognition: - A recognition system is used for identifying the respective computer character once the single character extraction has been done. Various distinct methods for recognition are now available.[4] 

Feature extraction: -Feature extraction functions in an equivalent way like the recognizers of the neural networks. Although, programmers should set the desired properties manually which they. The properties used in the identification allows more control in this approach. Due to the properties are not learned automatically, any system using this approach still requires considerably more time for the development than a neural network.[4] 

Optical character recognition: -Optical character recognition may be discovered earlier for creating blind people’s reading devices and technologies including telegraphy. Emanuel Goldberg produced a machine that read characters and converts them into standard telegraph codes in 1914. Simultaneously, Edmund Fournier d’Albe produced the Optophone, a handheld scanner that develops tones which are equivalent to characters or specific letters while moving over a printed page. Lately, in the 1920s and during the 1930s Emanuel Goldberg produced a “Statistical Machine” as called by him for exploring microfilm archives through making use of optical code recognition system. For the invention, he was permitted USA Patent number 1,838,389 in 1931. Then IBM took possession of the patent.[11] 

3.2 State of the art 

Generally, it is a difficult task for recognizing unrestricted handwritten content lines. The finest latest recognizers even lead to fewer recognition figures due to the difficulty of dividing cursive or overlapping words, merging with the requirement for the absorption of context details. Either by development in language modeling or by advanced preprocessing, nearly all current progress in the field has happened. Comparatively some work is done on the fundamental recognition process. Despite commonly known for imperfection certainly, mostly systems depend on the same concealed Markov models which have been used in speech and handwriting recognition for a long time.[10] 

Handwriting Recognition development since 2009 

The research group of Jurgen Schmidhuber at the Swiss AI Lab IDSIA that had achieved various international competition of handwriting have evolved the deep feedforward neural networks and the recurrent neural networks since 2009. Deprived of acquiring any previous knowledge of the three distinct languages that are French, Arabic, and Persian, Alex Graves et al. secured first place in the competition at the 2009 International Conference on Document Analysis and Recognition (ICDAR) of connected handwriting recognition specifically in the bi-directional and multi-dimensional long short-term memory (LSTM). Current GPU-based deep learning techniques for feedforward networks by Dan Ciresan and fellow worker come first in the ICDAR 2011 offline Chinese handwriting recognition competition; their neural networks also were the foremost artificial pattern recognizers for achieving the human-competitive performance on the famous MNIST handwritten digits problem and of Yann LeCun and fellow worker at NYU.[4] 

Modern Techniques 

Conventional methods emphasize on dividing single characters for recognition, whereas modern techniques emphasize on recognizing each character in a sectioned line of text. Specifically, they emphasize on the methods of machine learning that can learn visual attributes, evading the restricting feature engineering used before. State-of-the-art methods use convolutional networks for extracting visual attributes in place of various overlapping windows of a text line image that Recurrent Neural Network makes use of for producing character chances.[4] 

3.3 Case Studies 

3.3.1 Handwriting Recognition using Neural Network 

Artificial Intelligence can be said to act on the specific domain intelligently and as the environment perception. The neural network is one of the instruments of AI. The artificial computing and science experience for recognition of optical character has been enhanced by neural networks.  The English characters (A-Z, a-z), numerals (0-9), and special characters (#, $, %, &, *). can be recognized using the system. The paper is focusing to apply the system for free handwritten characters, numerals, and special characters. Through training the neural network sufficient times using a backpropagation algorithm, the recognition is done. The paper is aiming for developing the offline strategy for free handwritten characters.[5] 

3.3.2 Handwritten Character Recognition using Neural Networks 

Recognizing the characters in the given documents which are scanned and studying the results of changing the Artificial Neural Network models is the aim of this article. Pattern Recognition is almost done by neural networks during recent times. This article expresses the behaviors of various neural network models that are utilized in Optical Character Recognition. The neural network is extensively used by Optical Character Recognition. The number of concealed layers, epochs and hidden layer size is the specification that has been considered in this article. Multilayer feed-forward network with backpropagation has been used in this article. Some primary algorithms have been applied for characters normalizing, characters partition, and deskewing in preprocessing step.  For finding the accuracy of the corresponding neural network, various models of the neural network have been used and implemented the test set on each in the system.[1] The given application is useful for recognizing all English characters given in the form of an image as an input. The character image is given as an input to the presented system, which then will recognize input character that is given in image. The neural network recognizes and classifies characters. The vital purpose of this project is to recognize a specific character of type format using the Artificial Neural Network technique effectively. Their research will be supportive of the one who would be willing to research other scripts.[6]  

3.3.3 Neural Networks for Handwritten English Alphabet Recognition 

This article depicts the usage of neural networks for producing a system that recognizes handwritten English characters. In this method, each English alphabet is represented by binary values which are used as input to a simple feature extraction system, whose output is given to the neural network system. The system has been proposed and developed for recognizing handwritten English alphabets. Almost all English alphabets have experimented with testing with different Handwriting styles. 82.5% average accuracy has been achieved by the machine as stated from experimental results that are significant and may be implemented in some areas.[12] 

3.3.4 Diagonal Based Feature Extraction for Handwritten Alphabets Recognition System using Neural Network 

This article describes an off-line handwritten alphabetical character recognition system which uses various layer feed-forward neural network. For taking out the features of the handwritten alphabets a new method called diagonal based feature extraction is established. For training the neural network, 50 different datasets consisting of 26 alphabets described by diverse people each; For testing 570 distinct handwritten alphabetical characters. As compared to the systems utilizing the traditional vertical and horizontal procedure of quality extraction, the given recognition system executes effectively giving excessive recognition precision. For recognizing handwritten characters and obtaining constituted text type from handwritten documents, this system will be suitable.[6]  

3.3.5 Image preprocessing for optical character recognition using neural networks 

The prime aim of this paper is to use forward-feed neural networks and, create a practical and theoretical based on the printed text for optical character recognition preprocessing. As stated by the results of realized experiments, demonstration application was created, and its parameters were set.[6] Mostly all who work with the computer have to give input from the paper to the computer. There are multi ways to do that task. The intelligent way is to first scan the document and after which optical character recognition converts the scanned image into the text that can be edited. As stated in the paper, the neural networks can be applied satisfactorily to the preprocessing with many advances as compared to the standard image preprocessing methodologies.[13] 

3.3.6 Intelligent Systems for Off-line Handwritten Character Recognition: A Review 

In the field of recognition of the pattern and processing of the image, Handwriting Recognition has been always an important research area. There is also a high demand for Optical Character Recognition on handwritten documents. This article presents a thorough review of HCR works that are in existence during the last 10 years based on the technique of soft computing.[6] 

3.3.7 Fuzzy Based Handwritten Character Recognition System 

This article depicts a fuzzy technique for recognizing handwritten characters. For the recognition and depiction of fuzzy character, fuzzy logic and sets are used as ground. This article presents an algorithm that is based on fuzzy that firstly divides the character and then gives the possible characters which match with the input given using the fuzzy system and finally recognizes the character using the defuzzification system.[6] 

3.3.8 Recognition for Handwritten English Letters: A Review 

In the subject of Image processing, Character recognition is one of the most challenging and interesting research areas. In the last half-century, character recognition of English has been comprehensively studied. For character recognition, various techniques are used widely currently. Different application fields of digital document processing are digital library, postal addresses reading, entry of data, health insurance, tax forms, loans, verification of documents, bank deposit slips reading, cheques information extraction, credit card applications, etc. This article gives an outline of exploration done for the handwritten English letters’ recognition. There is no restraint on the style of writing in handwritten text. As a result of varied styles in human handwriting, angle variation, and letters shape and size it is strenuous to recognize Handwritten letters. Along with their execution, handwritten character recognition varied methods are considered in the article.[6] 

3.3.9 An Overview of Character Recognition Focused on Off-Line Handwriting 

Character Recognition has been comprehensively learned in the last 50 years and developed sufficiently for producing technology-driven applications. Today, the computational power that is flourishing quickly makes us able to implement the present Character Recognition methods and generates a growing demand for many prominent application domains that requires more advanced methods.[6] 

3.3.10 Offline Handwritten English Numerals Recognition using Correlation Method 

The author has presented the system which successfully recognizes the offline handwritten digits with greater accuracy than the works done previously in this article. Also, the handwritten number recognition systems were capable to recognize only single digits and the systems were not able to recognize multiple digits at a time. Thus, the author is trying to concentrate on efficiently performing partitioning for separating the digits.[6] 

 

3.4 Future Prospects 

Due to the methods of feature extraction like character geometry and gradient technique are utilized in the method does not categorize different languages characters, the method can be elongated for the classification of languages independently from the images of other languages with few changes. For attaining a greater accuracy rate, the filtration of divided characters can be done. By including few more features besides the prevailing ones, the neural network performance can be increased. By training the neural network with a greater number of images for testing, the classification ratio can be enlarged.[1] The handwriting recognition using a neural network's work can be taken forward for other language character recognition.  Text format can be obtained by using it for the conversion of the fax and the newspapers. For classifying, multiple Artificial Neural Network can be used for recognizing words, sentences, or paragraphs. Also, for reading postal addresses in the post office it can be used.[6] 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

4 Conclusions 

Handwriting systems can be used for recognizing plenty of different regional languages having distinct writing styles applying respective algorithms and strategies. Because of the strange character's presence or resemblance in shapes of various characters, it has emerged that handwritten character recognition becomes difficult. The characters are divided into single characters and for getting a clean image, the scanned image is pre-processed. Normalization, filtration is done with the usage of processing steps results in clean and noise-free outcome under pre-processing work. The system desired result with effectiveness can be achieved by governing algorithms with assessment, proper training, and other step-by-step procedure. For achieving finer English characters recognition output, some geometric and statistical features must be used.[6] Artificial Neural networks can be used in handwriting recognition which can be done using a feed-forward backpropagation algorithm that is applied to the collection of data at a specific time. Handwriting recognition can be done effectively by using proper methods with accuracy and method complexity. Handwriting recognition can be done by using feature extraction, clustering, matching of the pattern; but it can be said as per the research done that the higher accuracy, efficiency, and reliability can be achieved by using neural networks.[5]  

 

 

 

 

 

 

 

 

 

 

 

 

 

5 References 

[1] R. V Kalava, “Chapter. 1 INTRODUCTION 1. 1.,” no. 3, pp. 1–11, 1942. 

[2] N. Nain and S. Panwar, “Handwritten Text Recognition System Based onNeural Network,” J. Comput. Inf. Technol., vol. 2, no. 2, pp. 95–103, 2016. 

[3] O. H. Artificial, A. N. Types, C. Recent, and R. External, “Neural network,” pp. 1–10, 2020. 

[4] M. M. Mijwel, “Handwriting Recognition Methods,” no. September 2015, pp. 1–6, 2018. 

[5] S. S. Kharkar, S. S. Gadhari, S. S. Shrikhande, and P. D. K. Chitre, “Handwriting Recognition using Neural Network,” vol. 5, no. 4, pp. 1179–1181, 2017. 

[6] T. M. Breuel, “Handwritten character recognition using neural networks,” Handb. Neural Comput., no. March, 2004, doi: 10.1887/0750303123/b365c93. 

[7] T. L. Fine, “Feedforward Neural Network Methodology,” IEEE Trans. Neural Networks, vol. 12, no. 3, pp. 647–648, 2001, doi: 10.1109/TNN.2001.925573. 

[8] A. Mehta, “A Comprehensive Guide to Types of Neural Networks,” Digitalvidya, p. 2, 2019. 

[9] N. S. Behbahan, “A Method based on Decision Trees and Neural Network for Recognition of Farsi Handwritten Digits,” Case Stud. J., vol. 2, no. 2013–05, pp. 01–08, 2013. 

[10] M. Liwicki, A. Graves, and H. Bunke, “Chapter 2 Neural Networks for Handwriting Recognition,” pp. 5–6. 

[11] W. D. Ryan and D. H. Campbell, “Optical character recognition device,” Eur. Solid-State Circuits Conf., vol. 1976-Septe, p. 33, 1976, doi: 10.1109/ESSCIRC.1976.5469239. 

[12] Y. Perwej and A. Chaturvedi, “Neural Networks for Handwritten English Alphabet Recognition,” Int. J. Comput. Appl., vol. 20, no. 7, pp. 1–5, 2011, doi: 10.5120/2449-2824. 

[13] R. Jakša, “Image preprocessing for optical character recognition using neural networks,” pp. 1–11.