Moving Objects Segmentation Based On Isoperimetric Algorithm

In dynamic scene images segmentation there are relative motion between objective and camera. When the objective partition moves, the video device had also a corresponding movement. The changes caused by the moving objects and the moving background are mixed together. Too many segmented regions are produced. In such case how to extract the moving objects accurately in real-time has become the key point in this paper.The segmentation approaches based on spatio-temppral information is used in the paper. In the process we adopted the space clustering information in the current frame, like gray, texture, and other information. At the same time we also adopted the motion information in consecutive frame.In the aspect of spatial segmentation, image segmentation based on graph theory is used in the paper. The theory is a new image segmentation technique developed in recent years. We got the image segmentation result according to isoperimetric cutset criteria. The graph partitioning has been strongly influenced by the classic isoperimetric problem: for a fixed area, find the region with minimum perimeter. The isoperimetric algorithm effectively restrains over-segmentation. The theory allows us to use a solution to a linear system instead of solving an eigenvector problem. The running speed and the stability are increased. We get an accurate segmentation.In the aspect of temporal segmentation, we achieved through the motion information of the image sequence. The motion information performed that the same object in different frame was different. The motion detection technology in the interframe was required. The paper was adopted global motion estimation and compensation. We made 6 affine motion to motion models and global motion parameters were achieved by over-relaxation iteration. After we got the global motion parameters, we used bilinear interpolation to reconstruction the pixel brightness. By global motion compensation, we did differential operation between the current frame and the compensation frame, and obtained local moving region.For the difference image, we connected change points by the regional filled and the mathematical morphology. Changes points divided by the total number of pixels in the region as a comparative value. The comparative value compared with a threshold to determine the motion regions. In accordance with the scope of the changes in the differential image to merge the motion regions, we got the moving objects. The paper proved its effectiveness through simulation experiments.