A correlation coefficient based model to separate and classify noncursive (Grantha script) symbols

Abstract

Symbols are confined to documents either as isolated notations or hand-written texts with a number of notable features, however distinguishes, from other writing variations. This paper describes a method to separate and classify handwritten non-cursive symbols (of Grantha) from document images. This method use statistical correlation coefficient for separation and classification instead of recognizing the symbols. The model comprises of selection, separation of symbols and preprocessing steps, like normalization, skeletonization, and finally, the classification. The method employs bounding box algorithm for the location of script symbols in the document images. The efficiency of the method is, as such, it selects only the script symbols and excludes non-symbol components. In the proposed method, preprocessing steps makes the separated symbols suitable for classification. For experimental verification, 50 degraded document images of varying deteriorating complexities were tested. The resulting symbol classification rate (i.e., the proportion of symbols automatically classified) was obtained close to, ≈ 70%.