Video Supervoxels Using Partially Absorbing Random Walks

Supervoxels have been widely used as a preprocessing step to exploit object boundaries to improve the performance of video processing tasks. However, most of the traditional supervoxel algorithms do not perform well in regions with complex textures or weak boundaries. These methods may generate supervoxels with overlapping boundaries. In this paper, we present the novel video supervoxel generation algorithm using partially absorbing random walks to get more accurate supervoxels in these regions. Our spatial-temporal framework is introduced by making full use of the appearance and motion cues, which effectively exploits the temporal consistency in video sequence. Moreover, we build a novel Laplacian optimization structure using two adjacent frames to make our approach more efficient. Experimental results demonstrated that our method achieved better performance than the state-of-the-art supervoxel algorithms.

I. Introduction

A supervoxel is commonly defined as a subset of 3-D grids of video volume, which contains voxels that are similar in color, texture [13], etc. It is regarded as a perceptually meaningful atomic volume in a spatial–temporal domain. Similar to superpixels [33], supervoxels carry richer information such as spatial–temporal consistency, and they are often more manageable than simple voxels. Thus, supervoxels have been generally used for many video analysis and processing applications [19], [35], [36], such as video tracking, object recognition [32], motion analysis [31], and scene reconstruction [28], [34], [37]. A good supervoxel should own a regular shape with a uniform size and adhere well to the boundaries of object, which will improve the performance and decrease the computing cost of video processing applications.