关键词: 视频对象分割 非线性尺度空间 Total Variation Flow模型

Abstract:

An algorithm based on the nonlinear scale space theory is proposed in this paper for video object segmentation.Firstly,every frame in the video sequence is decomposed into a series of multi-scale images.Then,the video object is determined by the moving information from large-scale image sequences.At l ast,the precise boundaries of the moving objects are extracted according to the series of the multi-scale images of every frame in the sequence.Experime ntal results show that the algorithm can robustly extract moving objects from the complicated natural scene.

Key words: video object segmentation;nonlinear scale spaces;Total Variation Flow model