Recognition of handwritten Bengali characters: a novel multistage approach
Introduction
Recognition of Bengali characters is a difficult task. The main difficulties encountered are the extreme cursiveness of the written characters, even when they are separately written, the abundance of very similar symbols and a very large symbol set. An analysis of the Bengali character set reveals that there are some very distinct characteristics present in some symbols which are completely absent in others. Utilising these characteristic features can help the design of OCR systems by forming a multistage recognition scheme. This paper describes in detail what type of characteristic high-level features can be considered and how these features can help in boosting recognition performance. Since this is a difficult task, specific target oriented information is extremely important, but at the same time it is also vital to incorporate additional information from multiple classification to make the overall decisions more robust. Multiple expert decision combination has become a very attractive option in designing various pattern recognition and document analysis applications. The strength of these approaches lies in the way a majority consensus is either inherently or explicitly implemented in designing such frameworks (See [1], [2], [3], [4], [5], [6], [7], etc.). It is also known that the success of multiple expert solutions depends on the extent to which a priori information about the target task domain is exploited (see, for example, [8], [9]). In this paper, a recently reported novel multiple expert decision combination approach has been adopted to produce a very robust classification framework. This approach is based on a conventional majority voting scheme, but has an associated evaluation phase designed to function as an integral part of the decision combination framework. This evaluation phase evaluates the merits of a majority decision and makes the final combined decision more robust.
Section snippets
Analysis of the Bengali character set
Bengali character set consists of 49 classes. A Bengali handwritten database has been compiled. The characters were written separately and in this sense it is a compilation of hand-printed characters. The scanned resolution was 400×200 pixels. A GT-4000 Seiko Epson Colour Image Flat-bed Scanner was used for scanning. In the image files, a 1 indicates an absence of characters and a 0 indicates a presence, so the pixels were in reverse-video mode. However, no restrictions were imposed on the
Multistage recognition scheme
Fig. 4 outlines the multistage recognition scheme proposed in this paper. There are two distinct stages in this approach. In the first stage, Stage I, the five high-level features described in Table 1 are determined. Based on these features, the patterns are coarsely classified into seven different super-classes, each super-class representing multiple character classes. For example, referring to Table 2, 19 character classes are collectively expressed as Superclass 2 and 2 character classes are
Conclusion
A novel multistage recognition scheme for recognition of handwritten Bengali characters is presented. An investigation into the identification of some basic high-level features from the Bengali character set has demonstrated that detection of these features can help to form small sub-groups of characters and can act as a valuable source of secondary a priori information in devising classification strategies. This information has been exploited in designing a multiple stage recognition scheme
Acknowledgements
The authors gratefully acknowledge the support of the UK Engineering and Physical Sciences Research Council.
About the Author—A.F.R. RAHMAN graduated in Electronics from Bangladesh University of Engineering & Technology (BUET) in 1992. He subsequently undertook research in pattern recognition, studying the classification of Bengali handwritten characters, for which he was awarded the M.Sc. degree in 1994. He studied for the Ph.D. degree at the University of Kent in the UK and is now on the research staff of the University. Dr. Rahman is a member of the British Machine Vision Association and has
References (24)
- et al.
Machine-printed character recognition revisited: re-application of recent advances in handwritten character recognition research (special issue)
Document Image Process. Multimedia Environ. Image Vision Comput.
(1998) - et al.
An evaluation of multi-expert configurations for recognition of handwritten numerals
Pattern Recognition
(1998) - et al.
A new hybrid approach in combining multiple experts to recognise handwritten numerals
Pattern Recognition Lett.
(1997) - et al.
Recognition of handwritten numerals with multiple feature and multistage classifier
Pattern Recognition
(1995) - et al.
Performance comparison in hierarchical architectures for memory network pattern classifiers
Pattern Recognition Lett.
(1986) - et al.
Decision combination in multiple classifier systems
IEEE Trans. Pattern Anal. Mach. Intell.
(1994) - et al.
Methods of combining multiple classifiers with different features and their applications to text-independent speaker recognition
Pattern Recognition Artificial Intell.
(1997) - J. Kittler, A. Hojjatoleslami, T. Windeatt, Weighting factors in multiple expert fusion, Proceedings of British Machine...
- et al.
Application of majority voting to pattern recognition: an analysis of its behavior and performance
IEEE Trans. Pattern Anal. Mach. Intell.
(1997) - et al.
Enhancing multiple expert decision combination strategies through exploitation of a priori information sources
IEE Proc. Vision Image Signal Process.
(1999)
Cited by (64)
Handwritten Bangla character recognition using a soft computing paradigm embedded in two pass approach
2015, Pattern RecognitionCitation Excerpt :Multistage approaches require grouping of character classes with samples of similar shapes. In the work Rahman et al. [4], ten disjoint combinations of five script specific topological features are used for this. In the works of Bhattacharya et al. [7,21], character classes with samples showing high rate of mutual misclassification are grouped together.
A multifaceted evaluation of representation of graphemes for practically effective Bangla OCR
2024, International Journal on Document Analysis and RecognitionPipeline Enabling Zero-shot Classification for Bangla Handwritten Grapheme
2023, BLP 2023 - 1st Workshop on Bangla Language Processing, Proceedings of the WorkshopHand Written Digit Recognition using Multilayer Deep Convolutional Neural Network
2023, Proceedings - 3rd International Conference on Smart Technologies, Communication and Robotics 2023, STCR 2023A machine learning approach for Bengali handwritten vowel character recognition
2022, IAES International Journal of Artificial IntelligenceBangla Handwritten Character Recognition Method
2022, 2022 IEEE International IOT, Electronics and Mechatronics Conference, IEMTRONICS 2022
About the Author—A.F.R. RAHMAN graduated in Electronics from Bangladesh University of Engineering & Technology (BUET) in 1992. He subsequently undertook research in pattern recognition, studying the classification of Bengali handwritten characters, for which he was awarded the M.Sc. degree in 1994. He studied for the Ph.D. degree at the University of Kent in the UK and is now on the research staff of the University. Dr. Rahman is a member of the British Machine Vision Association and has published over 40 papers in the technical literature. His research interests are mainly in text recognition, with a special emphasis on the development of multiple expert decision combination structures, but he is also involved in document analysis and neural network processing.
About the Author—R. RAHMAN graduated in Biology from Dhaka University in Bangladesh in 1995. At present she is carrying out research at the University of Kent in the UK on computer analysis and assessment of stroke patients. Her research interests are mainly in static and dynamic drawing analysis, but she is also involved in pattern recognition and document understanding.
About the Author—MICHAEL FAIRHURST has been on the academic staff of the Electronic Engineering Laboratory at the University of Kent since 1972. He has been actively involved in various aspects of research in image analysis and computer vision, with a particular interest in computational architectures for image analysis and the implementation of high performance classification algorithms. Application areas of principal concern include handwritten text reading and document processing, security and biometrics, and medical image analysis.
Professor Fairhurst is a current member and past Chairman of the IEE Professional Group E4 on Image Processing and Vision, and has in the past been a member of the Professional Group Committee for Biomedical Engineering. He has been Chairman of several of the IEE International Series of Conferences on Image Processing and Applications, including IPA 99 held in July 1999. He has been a member of many Conference Organising and Programme Committees, including the most recent IWFHR Workshop, and is a member of the British Machine Vision Association. He is a member of the IT and Computing College of the EPSRC. He has published more than 200 papers in the technical literature and has authored an undergraduate textbook on computer vision.