Character Component Segmentation and Categorization of Machine Printed Text in Devanagari (Nepali) Script in Digital Image Processing

摘要

Extraction of core character with related components, exactness and consistency of extracted symbol determines accuracy of any Digital Image Processing Systems. Furthermore, pre-categorization of extracted symbols reduces lots of processing load at classification level. This paper proposes a core character and its component segmentation and categorization method of Machine Printed Text in Devanagari-Nepali Script. This paper also presents a method that extract modifier components which are not connected to core character. Here, Shirorekha or Dika or header line is considered as major component of segmentation and categorization. The proposed model removes the Shirorekha or Dika using horizontal projection profile on word and label the image of word to extract the objects as components. We have supplied one object has one label. Some character loose some property at removal of Shirorekha. Thus, we have reconstructed character for exactness and consistency of extracted object. We have used a set of structural layout - height width ratio, appearance position in word, presence of Shirorekha over the extracted symbol to categorize the extracted objects. We categorize extracted symbol into five categories - Non-dika character, regular character, conjuncts, upper modifiers and lower modifiers. The result obtained shows we have an accuracy of 98.26% to 100% as compared to previous method.