An HMM-MLP Hybrid Model for Cursive Script Recognition

Abstract.

This paper presents an HMM (Hidden Markov Model)-MLP (Multi-Layer Perceptron) hybrid model for recognising cursive script words. We adopt an explicit segmentation-based word level architecture to implement an HMM classifier. An efficient state transition model and a parameter re-estimation scheme are introduced to use non-scaled and non-normalised symbol vectors without having to label primitive vectors. This approach brings well-formed discrete signals for the variable state duration of the HMM. We also introduce a new probability measure as well as conventional schemes to combine the proposed HMMs and a general MLP. The main contributions of this model are a novel design of the segmentation-based variable length HMMs, and an efficient method of combining two distinct classifiers. Experiments have been conducted using the legal word database of CENPARMI with encouraging results.