Feature Extraction for Face Recognition using Edge Detection and Thresholding

Abstract

Face recognition is first and foremost step in video surveillance applications which include human behavioral analysis, event detection, border security and ATM banking. Most of the time, it is very difficult to get good facial features from the particular image frame and it often requires sophisticated algorithm for face identification and recognition. Robust face detection system is still a more challenging job because of complex environments including illumination changes, background clutter and occlusions. This article presents a novel feature extraction algorithm for face recognition using edge detection and thresholding. Initially, the incoming image is preprocessed to smoothen the image features and it is converted in to grayscale image to reduce the computational complexity of post processing steps. In feature extraction step, the image is completely iterated throughout the spatial coordinates and the edges are detected using thresholding technique. The optimum threshold for global thresholding is identified by calculating the maximum between-class variance in the given image. The extracted edge features are invariant under scale and illumination changes and thus it ensures the robust binary mask for face identification. Finally, the foreground features are obtained using morphological operations and the face is highlighted in subsequent incoming image frames. The proposed method can be deployed in public places such as malls, ATM centers and airports for security applications. Experimental results clearly indicate that the proposed approach works well under complex situations.