Automatic Recognition of Legal Amounts on Indian Bank Cheques

Abstract

Holistic-based approaches attempt to represent an entire handwritten word as an indivisible entity by representing it with feature representations. Despite the presence of various feature representations, it still remains a challenge to get the effective representation for Devanagari Legal amounts. In this paper, an attempt is made to represent legal amounts with histogram of oriented gradients (HOG) and local binary patterns (LBP) for their characterization. Thereafter, two fusion-based models are proposed. In the first model, HOG and LBP are fused at feature level and, in second, at decision level. Later, recognition is performed with the nearest neighbor and support vector machine classifiers. For corroboration of the efficacy of the proposed models several experiments have been conducted on ICDAR ' 11 Devanagari Legal amount dataset. Experimental results demonstrate that fusion based approaches are effective by achieving significant improvement in recognition accuracy as compared to individual feature representations and other contemporary approaches employed on the data set.