FFTSnake模型研究  

Research on FFT Snake Model

作　　者：李天庆[1] 张毅[1] 刘志[1] 胡东成[1] 

机构地区：[1]清华大学自动化系,北京100084

出　　处：《计算机工程》2005年第12期1-3,69,共4页Computer Engineering

基　　金：国家社会科学基金资助项目(99BTY004)

摘　　要：从3方面入手,对Snakes的外部力引入时空图光流特征,内部力引入傅立叶变换滤波平滑曲线,能量函数极小化方法改为法线方向寻优,从而使FFTSnake模型具有很好的实用价值。实验对比验证,在复杂背景下FFTSnake模型依然可以有效跟踪运动物体,此时普通Snakes模型必须做大量手动调整。而它同时也解决了帧差等视频分割方法难于识别、提取目标轮廓的问题。FFTSnake模型抓住了运动目标的图像特征,在简单的优化算法下依然可以得到较好的分割效果,并可以用于实时系统。For increasing the practicality of Snakes model, this paper attempts adding the angle of the grads of the spatio-temporal picture as the new outer force, replacing the tangential curve with Fourier transform filter as the inner force, and optimizing along the normal of the curve. By the experimentations, this new model can well track mobile object on the complex background. In the same condition, other Snake models are invalidated. On the other hand, the problem is also dissolved to recognize the edges by traditional video segmentation algorithms. FFT Snake model can track the mobile object rapidly because its optimizing algorithm is very simple. So, it can be used in real time system.

关 键 词：主动轮廓线模型 时空图 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]