Myoung-Ho Lee

Moving object segmentation in video sequences by user interaction and automatic object tracking

Abstract The new MPEG-4 video coding standard consists of object based coding schemes for multimedia and enables content based functionalities. In order to apply the MPEG-4 visual standard, each frame of video sequences should be represented in terms of video object planes (VOPs). Video objects in still pictures or video sequences should be first identified before the encoding process starts. It requires a prior decomposition of sequences into VOPs so that each VOP represents an object. This paper addresses a user-assisted video segmentation method for separating moving objects from the background in video sequences. The proposed method comprises of intra-frame and inter-frame segmentation modules. The intra-frame segmentation incorporates user interaction in defining a high level semantic object of interest to be segmented and detects precise object boundary. The inter-frame segmentation involves boundary and region tracking to capture temporal coherence of moving objects with accurate object boundary information. The proposed method exhibits a consistent and faithful segmentation performance for various types of video sequences.

New fast-search algorithm for block matching motion estimation using temporal and spatial correlation of motion vector

This paper introduces a new technique that reduces the search time and improves the motion estimation accuracy by using the high temporal and spatial correlation of a motion vector. Instead of using a fixed first-search point, as in previous search algorithms, the proposed method identifies a more accurate first search point through compensating the search area based on the temporal and spatial correlation of a motion vector. Accordingly, the proposed algorithm is based on the consistent directivity and center-biased distribution property of a motion vector. As a result, the performance of the motion estimation is improved and the total number of search points used to find the motion vector of the current block is significantly reduced. Simulation results showed that the PSNR values improved up to 3.6 dB, depending on the image sequence, and advanced on average by about 1.7 dB. The comparative results demonstrated that the performance of the proposed algorithm was better than those of other fast-search algorithms whether the image sequence contained fast or slow motion, and similar to the performance of a full-search (FS) algorithm. Furthermore, the performance of the proposed scheme produced a superior subjective picture quality compared with other fast-search algorithms.