Color And Object Outline Based Method In Video Segmentation

Video segmentation, storage and content-based retrieval become essential with the development of government affairs through network and web search. This issue is funded by the Jilin Province Scientific Development Project Content-based Multimedia Database and Its Application in Government Information Processing. Segmentation and clustering of the video sequences are the first step should be taken to build a content-based multimedia database. The conventional segmentation methods are based on histogram difference and pixel difference between the successive frame pairs. Pixel based method can detect the difference caused by any kind of gray variety ignoring the semantic contents set in group of continuous frames. Histogram based method takes the statistical gray levels difference as a measure segmenting the frames which in most cases shows poor results, especially when changes is gradual in color space such as the change caused by the motion of the camera.In this paper we propose a novel method solving the problems above. In order to compensate the affect with the pixel based method, we take into account the ratio of objects to their background. In cases when the background is static while the objects are moving or when camera is moving while the objects keep still, the changes between the ratio of objects to their background in the successive frame pairs appears indistinct. By clustering the difference feature resulted from the product of the two measures between successive frame pairs, we can detect the obvious change shot (usually called cut) and no obvious change shot (usually called dissolve), here we use fuzzy c-means method in clustering process. To select a representative frame from the video shot, average histogram is a conventional method, however, it results bad effects. Because the outlier frame may add noise to a shot in the same way affects the availability of the key frame selected through the video sequence. To avoid this kind of affect we choose the frame with maximum object ratio in background as the key frames sine more object ratio of a frame usually carries more information within a shot. I give the brief introduction of my proposing method in the following content. The first part is about the shot boundary detection. When the color distribution across the video sequence changes, a shot boundary may appear. Here we take the difference between successive frames as the main measure of detecting process. Conventional method compute the pixel difference between frame pairs pixel by pixel and sum the absolute difference throughout the frame pairs, however, the conventional method ignores the correlative character of the pixel intensity in the neighborhood across the frame pairs. Allowing for this feature, we take the point and their corresponding one together with its 8 neighbor points as the processing points. We compute the distance between the pixel within the window and it's corresponding 8 neighbor area pixels in the successive frame pairs. The minimum value is taken as the favorite distance between the pixel in one frame and it's 8-neighbor area in the corresponding frame. Throughout the frame pairs with the method and sum the absolute pixel distance we get the first measure for the video shot boundary detection. With only this measure, we can detect the shot boundary in a video sequence especially when there is obvious break between frames, however, it may not yield good result if we carry this method in unobvious shot boundary detection. The cause of this problem is usually the motion of not only the objects but also the camera panning. To supply the gap we take the difference of the changing difference of the objects outline as the adjoining feature to detect the shot boundary.In the situation the objects in a video move or the camera pans while the color distribution of the frame pairs keeps unobvious, the shot boundary detection mainly depends on the changing status of the object outlines and their position. To get the objects' ratio against the background we first outline the ed...