关键词: 关键词: 超像素分割, KD树, 混合高斯模型, 图割算法, 光流法

Abstract: Abstract: Video object segmentation is one of the most difficult problems in computer vision due to the factors like fast moving objects, cluttered backgrounds, arbitrary object appearance variation and shape deformation. In this paper we present a new unsupervised video segmentation algorithm based on multiple spatio-temporal feature representations. By the combination of saliency characteristics and other features obtained from pixels and superpixels, we design a coarse to fine-grained robust feature representation to represent each frame in a video sequence. First, we generate a set of superpixels to represent the foreground and background in order to improve computational efficiency and get segmentation results by graph-cut algorithm; then, the optical flow can be used to propagate information between adjacent frames, and the appearance of each superpixel can be updated by its non-local sptatio-temporally counterparts generated by the nearest neighboring searching method with the efficient K-dimension tree (KD-tree ) algorithm, so as to improve the robustness of the segmentation; after that, for segmentation results generated in superpixel-level, we construct a new Gaussian mixture model based on pixels to achieve pixel level refinement; finally, the algorithm calculate the saliency characteristics of each frame, as well as segmentation results generated by graph cut and Gaussian mixture model, to obtain a more accurate segmentation results by voting scheme. Extensive evaluations on the SegTrack dataset demonstrate the effectiveness of the proposed method, which performs favorably against some state-of-art methods.

Key words: Keywords：superpixel segmentation, KD-tree, gaussian mixture model, graph cut algorithm, optical flow