Scene tune recognition and detection in film videos

Abstract

It has been a research hotspot to retrieve quickly and accurately the film clips that reflect the fluctuation of audiences' mood according to individual user's needs. Detection of the film scene tune is an effective way to detect these clips. New features such as local motion share ratio,camera motion and shots similarity were proposed with the combination of film domain knowledge. Using the above-mentioned features and common features mentioned in other papers,a Bayesian classifier was utilized to classify shot types. According to the relationship between scene and emotion of audience,five scene tunes that could stimulate emotion of audiences were defined and detected based on the recognition of shot types. The experimental results show that the chosen features achieve good results. Compared with the results of other studies,both precision and recall rate of long shot and close-up shot are improved.