Pashto Characters Recognition Using Multi-Class Enabled Support Vector Machine

During the last two decades significant work has been reported in field of cursive language’s recognition especially, Arabic, Urdu and Persian languages. The unavailability such Pashto language is because of: absence a standard database research that ultimately acts as big barrier for community. slight change characters’ shape an additional challenge researchers. This paper presents efficient OCR system handwritten characters based on multi-class enabled support vector machine using manifold feature extraction techniques. These techniques include, tools zoning extractor, discrete cosine transform, wavelet Gabor filters histogram oriented gradients. A hybrid map developed by combining maps. performed developing medium-sized dataset encapsulate 200 samples each 44 language. Recognition results are generated proposed model map. An overall accuracy rates 63.30%, 65.13%, 68.55%, 68.28%, 67.02% 83% technique, HoGs, filter, DCT, DWT maps respectively. Applicability also tested comparing its with convolution neural network model. network-based rate 81.02% smaller than machine. highest SVM reflects applicability