Unsupervised feature based key-frame extraction towards face recognition

A convenient and most effective method of querying a video database for robust face recognition is by using keyframes extracted from the image sequence. In this paper we present a clustering based approach that bypasses the need for shot detection or segmentation, to extract the key-frames from the video using the local features, for the purpose of face recognition. Local features which are insensitive to noise, displacement, scale, rotation and illumination, are extracted from arbitrary points on the images based on Speeded Up Robust Feature (SURF) algorithm. The frames are then clustered using sequential K-means algorithm. A representative frame from each cluster called the key-frame is then determined for subsequent use in video based face recognition. The proposed method has been demonstrated with experimental results obtained using Honda/UCSD (name of a standard database available for face recognition research) dataset 1.