Shot is the fundamental unit of manipulation in video. Success shot detection is the basis of any content-based video searching system. In this paper, we propose a new general shot detection technique based on 3D hints extracted from 2D images and dynamic threshold selection. as compared to other techniques based on global visual features such as global color histogram, the experimental results at a real scale show that our technique is more accurate, and gives a more reliable behavior with respect to annoying effects such as speed camera movements. Indeed, by extracting 2D and 3D information contained in 2D images, our technique allows a deeper underrating of images as they characterize the geometric nature of a scene. The further work we have been undertaking shows that 2D and 3D hints so extracted allow us to detect other transitions than cut, camera movements such as panning, fade in/out, and even object movements and using high DC frequencies our technique can also be easily combined with compressed video.
展开▼
