Real Time Hand Gesture Recognition by Skin Color Detection for American Sign Language

This paper demonstrates a system that can translate American Sign Language into text from real time video with computer vision based approach. This system has two parts, Skin Color Detection (SCD) and Hand Gesture Recognition (HGR). Skin pixels are detected from the image by applying some heuristic rules and using an Artificial Neural Network (ANN). The ANN is trained by H, S, V component values from HSV color space along with five texture features. Automated threshold value for SCD is achieved by plotting Q component value from YIQ color space vs the result from ANN for each pixel and applying K-means Clustering on it. XOR operation is used to find out a standstill hand image from the output images of SCD system. For HGR another ANN is used. Features are extracted using two algorithms. They are fingertip finding algorithm (a combination of convex hull and K curvature) and pixel segmentation.