Implementation and Development of Soft Computing Methods and Their Performance Analysis for Hand Written Devanagari Character Recognition

Abstract

The problem consists with the study, development, implementation and analysis of the soft computing methods for the recognition of handwritten Devanagari characters. There are various techniques proposed in the literature for the character recognition of handwritten characters. These techniques are mostly using the offline learning mechanism of the training and testing. Every method has its pros and cons. The requirement is to analyze the performance of these methods for the given training set of handwritten characters with different forms of learning. The analysis for the performance of these method reflects about the possibility of exploring the new techniques and approaches for obtaining the more efficient and prominent result for the recognition purpose. newlineThe research focuses on devnagari handwritten character recognition. The detail review of research carried out in the devnagari HCR field is provided in chapter 2. The dataset is created by collecting the samples of handwritten devnagari characters and documents. The size of the character sets for devnagari scripts is very large. It contains vowels, consonants, modifiers and compound characters which marks it difficult for recognition. newlineThe handwritten character recognition system includes preprocessing, segmentation, feature extraction and character recognition. Chapter 4 gives the experimental results of segmentation of handwritten documents. In the current research projection profile, run length and bounding box methods are used for segmentation of lines, words and characters. The projection based method gives 92% accuracy for line segmentation and 98% accuracy for word segmentation. newlineChapter 3 provides details about the different techniques that can be used to recognize handwritten characters. The neural network based approach is proposed for handwritten devnagari character recognition in the current research. Neural network technique is used because it can solve problem with several constraints. Four types of neural networks i.e. feed forward, radial basis