Detecting shot transitions based on video content

Detection of scene transition is the first step on video segmentation, indexing and analysis. Although scene classification by human can be performed with visual or sonorous attributes at the same time, machine automatic classification usually relies on feature extraction of main visual characteristics. The use of color, shape, digital sound processing and voice signal altogether are investigated in this work. The color detection is based on the color histogram and shape detection is based on edge map histogram. Sound characteristics are resolved with the extraction of seven characteristics: short time average energy, zero-crossing rate, energy band ratio, delta spectral magnitude, root mean square of square sum of signals, high sounds and low value characteristics ratios. A Bayesian network is used on the decision for the transition. Finally, a new form of grouping frames is proposed. The results of the proposed method are summarized to show its efficiency.