With the advance of multimedia technology, we can use the varaiety of data like image, video, speech, etc. These data are very huge in their size. Especially the size of the video data is extraordinary. So, we need the method with which we can manipulate these data efficiently.
To be stored and accessed efficiently, the video data must be segmented into shots. It can be done with the scene change detection. In previous approaches, they used global or local threshold to detect the scene change, and the threshold was decided heuristically. With the heuristic decision of the threshold, the scene change is detected well in common case, but in some special case, like in the scene with very large motion it works poorly. In this thesis, we present an effective scene change detection algorithm using perceptron. With perceptron learning algorithm, the perceptron is trained to classify the video frame into scene change class or non-scene change class. With this method, we can detect scene change automatically without deciding the threshold. And the detection ratio was enhanced.
