Efficient Object Segmentation Using Background Estimation for H.264 Video

Abstract

H.264 is an up-to-date video coding standard with wide applications. In order to improve the segmentation efficiency for H.264 video, a novel object segmentation approach using background estimation is proposed in this paper. Firstly the preprocessing including spatiotemporal filtering and motion field accumulation is used to remove the noisy motion vectors and obtain the dense motion field. Then the inter-frame coding modes are applied to estimate the background region, which can reduce the processing time for the subsequent global motion compensation. Finally depending on the residual motion field, the hypothesis testing is exploited to distinguish between moving object region and background region. Experimental results for several video sequences demonstrate high processing speed and good segmentation quality. The proposed approach can effectively extract the objects from H.264 video by using the estimated background.