Research On Algorithm Of Moving Object Segmentation In Video Sequences

With the rapid development of information technology and Internet,multimedia applications in people's lives are more and more. Research on multimedia information processing technology developes progressively. Efficient compression coding method is especially important for large amount of multimedia information data. As an important part of MPEG-4 video coding standard, object-oriented video coding technology is the key to improve the coding efficiency, while an efficient video object segmentation algorithm is the basis for object-oriented coding. Video segmentation technique also has a wide range of applications in image processing, computer vision and other fields. Accuracy of segmentation result has a direct impact on the the effectiveness of follow-up work. It is of great significance.Domestic and foreign scholars have done a great deal of research work on segmentation of moving object in video sequences and put forward a number of segmentation methods. But there is not yet a universal, reliable automatic segmentation algorithm. This thesis presents a simple and fast video object segmentation method for low bit rate multimedia video sequences based on the research and analysis of the classical segmentation method. First we get difference image by an improved accumulated frame difference method. Get frame difference of both the adjacent image and alternate image, and then take "and" operation on the two results. Difference image is obtained by accumulating the difference results. At the same time we get the gradient of fame image whose moving objects are segmented by Canny edge detection. The difference image includes both moving objects and exposed background. In order to eliminate the background to obtain the template only including moving object, we get the edge template by taking "and" operation on difference image and gradient image, and then have Hough transform on the edge template to obtain continuous edge of moving object. Through mathematical morphology process and twice scanning we get moving object mask. At last the moving object region is filled with original values. The algorithm takes advantage of the correlation between each frame, increase the utilization of the image and obtain more precise edge of the moving object.This algorithm is a spatio-temporal combination segmentation algorithm. It uses the frame-difference in time-domain and edge information in space-domain. The algorithm is simple, fast and accurate. The experiments show that this algorithm can get a good result for low bit-rate broadcast video sequences.