Automatic extraction, segmentation and recognition of multi-font Indian Pincode

摘要

This paper proposes a new algorithm for Pincode script identification (PSI) that would detect Pincode in the address image of the envelope. The developed system differentiates the scripts of English alphabets and numerals. A new method for normalisation of extracted chain code features is proposed. The extracted numerals of various font types (multi-font) were recognised using multi layer perceptron (MLP) with back propagation (BP) algorithm and Naive Bayes classifier (NBC). The recognised Pincode was used for automating the sorting process. The experimental results of recognition of isolated Pincode were discussed. Recognition accuracy of 95% for MLP-BP and 93.8% for NBC was obtained. The performance comparison of MLP-BP and NBC are discussed in this paper.