Automatic video stream indexing and retrieving based on face detection using wavelet transformation

Abstract

The increased availability and usage of on-line digital video has created a need for automated video content analysis techniques including indexing and retrieving. This paper presents a proposed simple new technique to detect human face and retrieve shot/shots from video stream, which it has. If the input frame has a human face, then the detection and matching is the goal to detect the shot/shots which it has this face in video stream. This proposed simple technique depends on skin color detection on shots of video with more accuracy and faster than the previous techniques. The matching in this proposed technique presents that shot/shots detected from the video, which have the same faces dependent on input face frame don't affected by dim face in the frames, face expressions, shapes, have glasses or not, shadows, or rotation. Experimental results on movies, clips, TV shows, news programs and football games demonstrate the effectiveness of our face detection algorithm that leads 97%, and 96% for retrieving shots in precision ratio on the average.