Hand detection in still images

Abstract

Hand detection has been an active area of research in the past few years due to its potential use in gesture recognition and human-computer interaction. In spite of the advancements made in this area, hand detection in static images still remains a challenge due to the high exibility and shape variations of the articulated hand. In this work, we propose a method based on region proposals generated by skin segmentation and classi cation on the basis of extracted features. Speci cally, the approach involves converting the RGB image into hybrid HCgCr colour space and then segmenting it into skin and non-skin regions using K-means clustering. The binary image obtained is smoothed by using morphological operations. After removing facial regions, we get region proposals for further processing. Various shape, colour and texture based features are extracted from these region proposals. These features include histogram of oriented gradients (HOG), dense colour histogram (DCH), gist and four-patch local binary pattern (FPLBP). Finally, features extracted from each region proposal are fed into a trained SVM classi er which gives a hand or a non-hand label. The dataset used is a combination of Oxford hand dataset, NUS hand posture I and NUS hand posture II datasets.