Zone Based Relative Density Feature Extraction Algorithm for Unconstrained Handwritten Numeral Recognition

The recognition of handwritten digit recognition has been a challenging problem among the researchers for few decades. This paper proposes a relative density feature extraction algorithm for recognizing unconstrained single connected handwritten numerals independent of the languages. The proposed method consists of four phases, namely, image enhancement (dilation), representation (zone based), feature extraction (relative density) and recognition (minimum distance classifier). The handwritten numerals must be enhanced with dilation, in order to connect the broken digits. After enhancement, the dilated binary images can be represented as a mid-point aspect ratio class interval values. The minimum distance classifier technique has been used to recognize the given numerals. The method yielded a satisfactory recognition rate of 92.85%, 99.28%, 98.95%, 98.72%, 99.29%, and 99.48% for Latin, Assamese, Devanagari, Manipuri, Malayalam and Oriya Handwritten Numerals respectively.