Threshold based segmentation using median filter for sign language recognition system

With the growing capacity of computer vision technology, there is more interest being directed towards the automatic recognition of sign language based solely on image sequences or videos. In this paper a well-built segmentation process is developed which helps to promote a better vision-based sign language recognition system. Segmentation is a challenging problem, and its accuracy implies more in the system construction. The method proposed over here is incorporated for object annotation purpose. Its main purpose is to filter the image with median filter and fused with global thresholding method to produce better segmented results. This is more reliable segmentation during hand shape change and hand crossing. This is a two pass scheme in which the image is filtered and then segmented with Otsu's method which helps in fast segmentation. The approach is demonstrated with comparison of conventional methods of filters and segmentation process. Four filters with three segmentation methods are experimented over here. This greatly facilitates the image localization process which is important for the purpose of recognition.