Hand Gesture Identification using Preprocessing, Background Subtraction and Segmentation Techniques

Hand Gestures can be identified as the most natural way for Human Computer Interaction as they impersonate how humans interact with each other. In addition to HCI they are used in various applications such as remote control, robot control, human computer interaction, military application and sign language identification. Hand gesture identification is usually implemented in three phases-hand gesture segmentation, feature extraction and gesture classification. In this paper we have compared various filtering, background subtraction and edge detection techniques for the purpose of gesture segmentation. Initially the hand gestures are captured by the camera and those images are preprocessed using the Non Local Mean Filter (NLMF), Contra harmonic Mean Filter (CMF), Spatial Filter (SF) and Temporal Median Filter (TMF). These filter performances are compared using Signal to Noise Ratio and the Mean Absolute Error. The background is then subtracted using the Gaussian Mixture Model (GMM) and their results are compared with Adaptive Background Mixture Model (ABMM). Final edge detection is executed using the Improved Global Swarm Optimization based Canny Edge Detection (IGSOCED) and these results matched with that of Sobel and Laplacian Edge Detection (LED) techniques using Root Mean Square Error values. Thus the final system eliminates the noise from the input images and extracts the edges in a more effective manner and these segmented images can be further used for feature extraction and hand gesture recognition.