A performance evaluation of NSHP-HMM based on conditional ZONE observation probabilities application to offline handwriting word recognition

The two-dimensional approach based on Non-Symmetric Half-Plane Hidden Markov Model (NSHP-HMM) has been successfully applied to the area of off-line handwriting recognition. A new version of NSHP-HMM model based on conditional ZONE observation probabilities was recently introduced. This new version, called NSHP^Z-HMM, provides an optimal solution to combine the effectiveness of 2-D modeling by NSHP-HMM with a zoning-based appropriate pattern representation. The contribution of this paper is the use of NSHP^Z-HMM based classifier for the recognition of handwritten words. In the experimental tests, we compare the performance of two feature extraction methods with and without K-means clustering algorithm. Three handwritten databases have been used to evaluate the proposed approach. Preliminary results are promising.