Hybrid HMM/ANN Models for Improving Offline Handwritten Text Recognition

Varidhi N.K.; Savitha C.K.; Prajna M.R.; Ujwal U.J.

doi:10.37591/joipprp.v3i3.836

Abstract

AbstractIn this approach, it proposes the use of hybrid Hidden Markov Model (HMM)/Artificial Neural Network (ANN) models for recognizing the unconstrained offline handwritten texts. Handwritten image normalization from a scanned image includes several steps, usually it begin with image cleaning, page skew correction, and line detection. For handwritten text line image several pre-processing steps to reduce variation in writing style are performed like slope and slant removal and character size normalization. The structural part has been modeled with Markov chains, and a Multilayer Perceptron (MLP) is used to estimate the emission probabilities. This approach presents a new technique to remove slope and slant from the handwritten text and also to normalize the size of text images with supervised learning methods. Slope correction and size normalizations are achieved by classifying the local extrema of text contours with MLP. With the help of Artificial Neural Network slant is removed in a nonuniform way. Experiments have been conducted on many offline handwritten text lines and the recognition rates are achieved. Keywords: Handwriting recognition, offline handwriting, HMM, hybrid HMM/ANN, neural networks, image normalization, multilayer perceptronCite this ArticleVaridhi NK, Savitha CK, Prajna MR et al. Hybrid HMM/ANN Models for Improving Offline Handwritten Text Recognition, Journal of Image Processing & Pattern Recognition Progress. 2016; 3(3): 16–23p.