Recognizing Bangla Handwritten Numerals: A Hybrid Model

Images of handwritten are non-identical to machine-printed ones. Orientation, distortion, writing style, similarity among digits and capturing angle of the image bring the confusion to detection. To ease the problem of human-computer interaction for handwritten digits, a hybrid model has been proposed in this paper. The model classifies digit image applying global, local and geometric features. To read an image more crystalline, the procedure is introduced with image blurring, otsu's binarization and thinning. After preprocessing, features are extracted including thin based local binary pattern, radon transform and radon cumulative distribution transform (RCDT), junction and endpoint. Tangent feature is extracted from junction and endpoint. Then all these features are scaled by min-max normalization. In the proposed architecture, the SVM classifier has been used. This technique has shown its supremacy regarding recognizing Bangla handwritten digits by achieving 97.25% accuracy.