Word Spotting of Handwritten Hindi Scripts by Circular Histogram of Oriented Displacement (CHOD) Features

This paper presents a segmentation-based word spotting technique for handwritten Hindi scripts using newly proposed shape descriptor called circular histogram of oriented displacement (CHOD). A word spotting model is developed by training multi-layer perceptron (MLP) with CHOD features. Metrics of evaluation used are k-precision (kPr) and mean average precision (MAP). The proposed technique has been evaluated on two datasets consisting of segmented handwritten Hindi (Devanagari) word images and has posted very good performance. This is an indication of CHOD features having good discriminative powers. The proposed technique has been compared with other techniques for the same datasets and found to compare very well.