Bengali Sign Language Recognition using dynamic skin calibration and geometric hashing

In this paper, we propose a method for Bengali Sign Language Recognition based on skin segmentation and geometric hashing. The skin area is obtained using a combination of dynamic color-based skin thresholding and mean shift segmentation of the original image. A novel feature extraction algorithm is introduced which tries to identify the hand by placing points at regular intervals along the perimeter of the hand blob. A novel dataset of 1147 images is also prepared for the task of training a hash table map with geometric co-ordinates of the feature points. The method is built to recognize static hand signs of 37 Bengali alphabets. Conducting tests on two sets of 37 signs, with varying the precision of feature points taken on each test, yielded an overall recognition rate of 51.35%.