Off-line cursive handwriting segmentation of isolated words based on contour analysis

Cursive handwriting is writing style acquainted by nicely link between adjacent characters. Nowadays, cursive handwriting recognition is widely used in various applications including bank check processing and automatic address reading. The objective of such handwriting recognition is to realize a ma...

全面介绍

Saved in:
书目详细资料
主要作者: Kurniawan, Fajri
格式: Thesis
出版: 2010
主题:
标签: 添加标签
没有标签, 成为第一个标记此记录!
实物特征
总结:Cursive handwriting is writing style acquainted by nicely link between adjacent characters. Nowadays, cursive handwriting recognition is widely used in various applications including bank check processing and automatic address reading. The objective of such handwriting recognition is to realize a machine that is able to understand the meaning of cursive handwriting. In the literature, cursive handwriting can be recognized based on character or whole-word recognition. Character-based recognition has advantage over whole-word recognition because the vocabulary can be dynamically defined and adjusted without the need of word training. However, the challenge in cursive handwriting segmentation is how to decompose sequence characters which need to be isolated. In addition, segmentation becomes more problematic in touching and overlapping cases. The aim of this research is to design off-line cursive handwriting segmentation using heuristic approach based on contour analysis along with segmentation point validation and touching character segmentation. Prospective segmentation points are determined by considering minimum number of cutting strokes and vertical projection. Heuristic segmentation determines whether the segmentation points are correct or incorrect. Thus, artificial neural network is adopted to validate each segmentation point. In this regard, the neural network is trained with two segmentation point classes i.e. correct and incorrect segmentation points. Afterward, segmentation based on self-organizing map (SOM) is applied to enhance segmentation accuracy. The experimental results show that neural validation and SOM segmentation have improved the segmentation accuracy. This is proven by the average segmentation error rate achieved which is 2.92% for the proposed segmentation method and 3.76% for the existing method, Enhanced Heuristic Segmentation (EHS).