一种基于非线性尺度空间的视频对象分割方法

A Video Object Segmentation Algorithm Based on the Nonlinear Scale Space Theory

本文提出一种基于非线性尺度空间理论的视频对象分割方法。该方法首先利用非线性尺度空间理论对视频序列中的每一帧图像做多尺度分解,然后在由大尺度图像组成的序列中利用运动信息确定运动对象,最后根据每一帧图像的多尺度图像确定对象的准确边界。实验结果表明,该方法能够在复杂的自然场景中精确地分割出视频对象,具有较强的抗干扰能力。

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 last,the precise boundaries of the moving objects are extracted according to the series of the multi-scale images of every frame in the sequence.Experimental results show that the algorithm can robustly extract moving objects from the complicated natural scene.