Color feature based video content extraction and its application for poster generation with relevance feedback

Poster that represents a video has many applications in internet and media world. In this paper, a method is proposed to generate poster based on color feature and relevance feedback. Color feature is used because it is more closely to human perception and it is easy to extract, match and also effective for indexing. In our method as color feature of a frame in a video, HSV color space and Color Moment are used. To calculate similar frames i.e. to detect shot, Euclidean distance between the features of frames are compared with a predefined threshold value. One frame is selected from each shot and stored with Temporally Occurrence Value (TOV) of the Frame. As a video can have similar shots in it but in different times, so the similar frames from the similar shots of different times are detected and one frame is selected from them. Then this selected frame is stored with Permanent Occurrence Value (POV) of the Frame by replacing its TOV and discarding others of them from storage. Finally, five frames with maximum POV are selected to generate poster layout and ranked the posters by their relevance feedback. The experiments show the robust result than existing methods.