Hand part segmentations in hand mask of egocentric images using Distance Transformation Map and SVM Classifier

Forearm and palm segmentation is a crucial element in the hand-pose estimation and mobility of the arms and hands estimation. However, hand parts separations (i.e, forearm and palm) from egocentric images have less been explored. In this study, we propose a novel hand forearm and palm segmentation method using the distance transformation map and an SVM classifier. First, we use a distance transformation map of the hand mask to find circles inscribing on the hand mask. These circles are vectored and construct an SVM classifier to predict the correct circle for approximating the palm region. Based on the predicted palm area, we propose a method for forearm and palm segmentations. The proposed method is evaluated on the rehabilitation dataset which includes egocentric hand mask images extracted from the hand rehabilitation exercise egocentric images. The results show that the proposed method successfully segments hand and forearm with high accuracy and requires low computational time.