M-Band Packet Wavelet Farsi Handwriting Word Recognition

Farsi Handwriting Word Recognition (FHWR), especially those with different orientation and scale changes as well as different handwriting style, is a challenging and important problem in document image analysis. This paper proposes a holistic FHWR scheme using local features of M-Band packet wavelet. The rotation and scale invariant local feature extraction for a given word image involves applying a polar transform to eliminate the rotation and scale effects, but at same time produce a row shifted polar image, which is then passed to an M-Band wavelet transform with row shift invariant to eliminate the row shift effects. So, the output M-Band wavelet coefficients are rotation and scale invariant. A local feature vector extracted from each subband of M-Band wavelet coefficients is constructed for rotation and scale invariant FHWR. In the experiments, we employed a Mahalanobis classifier to recognition a set of 224 distinct Farsi words selected from different persons with different style of writing. The experimental results, based on different handwriting style testing data sets for images with different orientations and scales, show that the proposed classification scheme using M-Band wavelet signatures outperforms other holistic handwriting word recognition methods, its overall accuracy rate for joint rotation and scale invariance being 95.8 percent, demonstrating that the extracted local energy features are effective rotation and scale invariant features.