Bangla Sign Language Recognition with Skin Segmentation and Binary Masking

The barrier between the hearing and speech impaired and society has been bridged through sign language for ages. Hereby, we present a system that recognizes Bangla Sign Language (BdSL) through the segmentation of hand gestures, binary masking and SVM classification. In the system, the features of different notations have been carefully distinguished and extracted to improve accuracy compared to the existing work. Through conversion into YCbCr color space and Fuzzy C-means clustering, a binary image is produced in the system which is later undergone binary masks designed for particular skin segmentations. Since YCbCr color space provides an effective use of chrominance information for modeling the human skin color, this color space has been used to pre-process the images. The dataset used in this system has been created and curated to perform under the manual feature extraction process. Extracting about 18 different features of the hand gestures of 10 alphabets, they are compared, classified and trained using a multiclass SVM classifier. Multiclass SVM classifier is extensively used for classification of high dimensional spaces and also effective for a smaller dataset. Using this methodology, the system can achieve an accuracy of 99.8% for static hand poses.