IMAGE SEGMENTATION IN VIDEO SEQUENCES: AN APPROACH BASED ON ONE CUT AND DEPTH MAP EXTRACTION by Yan Kuan Thesis Supervisor: Assistant Professor Liming Zhang Master of Science in Software Engineering Image segmentation is the technologies and processes that to divide the image into a number of regions with specific, unique properties, and made the object of interests. It is a key step from the image processing to the image analysis. There are hundreds of methods for image segmentation, for example, Graph Cut and Grab Cut. We aim to implement the image segmentation in a video sequence. The proposed method extract the depth maps from the video sequence based on motion estimation. The more efficient way of one cut algorithm is used in our approach. The traditional way of image segmentation includes the user interaction, which is not applied with our approach. We use depth map to generate foreground mask and background mask to implement automatic segmentation. In this thesis, we will introduce our whole approach about how to complete image segmentation in video sequences based on one cut and depth map extraction. We use depth map extraction based on motion estimation to obtain depth maps for every frames. In our approach we try to generate the foreground masks and background masks based on the depth maps automatically, in order to eliminate the users interaction. The one cut algorithm will complete the image segmentation for each frame based on the original pictures and the masks we insert to it. Finally we can get the segmentation images for every frame in a video sequences. There are many examples presented here, to demonstrate our approach from different aspects, such as time efficiency, segmentation effects, performance stability and so on.