Bangla handwritten characters recognition by using distance-based segmentation and histogram oriented gradients

Bangla Handwritten character recognition acquired a considerable attention in many research areas such as computer vision and image processing for its remarkable applications. In this regard, a Bangla handwritten character recognition method is proposed in this paper. The key challenges of this work are line segmentation, word segmentation, and character segmentation. For that in this paper, distance-based segmentation (DBS) method is used to segment the sentence, word, and character individually. To implement the DBS method efficiently, initially, the input Bangla document is pre-processed to resize and eliminate the noise. After that, adaptive thresholding method is utilized to remove the shadows from the input image. Furthermore, the proposed DBS method is applied to the processed image to segment the sentences, words, and characters. To segment words from a sentence, firstly segment the lines from the document. Based on these lines, the words in each line are segmented. Finally, characters are segmented from individual words. These segmented characters are used as ROI to extract the features and send to SVM to classify. To evaluate the performance of the proposed method manuscripts of Rabindranath Tagore and different peoples Bangla handwritten documents are considered.