The Joint Target Segmentation By SeamSeg And Superpixel Matching

Video segmentation technology is one of the hot issues in computer vision and graphic image processing.Its research purpose is to segment the foreground objects with significant characteristics from the complex background in videos.Therefore,video segmentation is widely used in the fields of behavior recognition,target monitoring and medical diagnosis.In practical application,due to the complexity of the background and the deformation,cover and luminous intensity change of the foreground target,the accuracy and robustness of video segmentation face severe challenges.In order to solve the problem above,this paper proposes a joint segmentation method based on Seam Seg and Superpixel matching.The main research results of this paper are as follows:(1)Because the complexity and diversity of video scene and shooting conditions in foreground target segmentation,the workload will be heavy and the algorithm will take a long time,if we segment the video frame directly.This method intends to use the region of interest to preprocess the target video and select the region around the target contour,which can greatly improve the validity of video frame pixel data and thus improve the robustness of the algorithm.(2)A joint segmentation method by Seam Seg and Superpixel matching was proposed.By using the foreground target contour that provided in the first frame of video and its manual label,we establish space-time relationship of video sequence with Seam Seg,and draw the target's rough outline by the above method.Then,The video frame is divided by SLIC superpixel segmentation,and the contour is modified according to the relationship between the semantic superpixel block formed by the mapping of neighborhood superpixel set strategy and the initial target contour,so as to obtain the contour edge with higher precision.(3)A matching method based on energy minimization is presented,we calculate the results of superpixel and Seam Seg segmentation method by combining the target segmentation region of the previous frame and image features such as color,center of gravity,shape and perimeter et al.Then we use the edges of the minimal energy gradient as a segmentation result.The final segmentation result is obtained by combining all local modified contours.We could maximum guarantee future target in complex background is still robust and accurate by these measures.