Re-evaluating chain-code as features for Bangla script

Most of the characters in Bangla script have similar shapes. Since chain-code is one kind of shape descriptor, we argue that it would not be able to distinguish between two similar characters and hence not be able to provide good recognition rate. Even though, the chain-code has widely been used as feature for Bangla script, none of the literature has considered the fact that chain-code may not be compatible with the script itself. We assume that chain-code cannot provide the variation necessary to describe the contours of similar characters, especially which are almost identical in the first place. We validated our proposal through a statistical test called one-way Analysis of Variance (ANOVA), to verify whether chain-codes of two similar looking characters are truly different. The results substantiate our assumption, suggesting that chain-code based features may not be the best features for Bangla character recognition.