Considerations for scene change detection method:

• Scene change detection should ignore frame-to-frame changes caused by camera motion, illumination changes, etc.
• Intensity histogram representation of a frame is enough for scene change detection
• Shot type classification; five basic gradations: close up, close shot, medium shot, full shot and long shot.
• Intercutting and parallel blending
• False cuts are harmless
• Misdetection of an actual cut is costly

Detecting scene changes is desirable in order to reduce processing time.

Shot detection scheme:

Input: bit stream of MPEG video which consists of three types of coded frames: intraframes(I), predicted frames(P) and interpolated bi-directional frames(B).
Step 1: Extract I frames from the encoded bit stream;
Step 2: Compute intensity, row and column histograms.
Step 3: Compare the current I frame intensity, row, and column histograms with those of the previous I frame. Chi-square statistic test is used to do the comparison.