A haar wavelet-based zoning for offline Arabic handwritten character recognition

Abstract EN

Due to the nature of handwriting with high degree of variability and imprecision, obtaining features that represent characters is a difficult task.

In this research, a features extraction method for handwritten Arabic Character recognition is investigated.

Its major goal is to maximize the recognition rate with the least amount of elements.

This method compute the 1 level Haar Wavelet Transform for Binary character image, then divide the Wavelet space into 8 Zones, for each Zone, three features have been extracted: mean, standard division, and skewness.

The Recognition have been done using Mahalanobis distance.

The proposed method provides good recognition accuracy of 73 % for handwritten characters even with fewer train samples