GVF Snake模型中初始轮廓线设置算法的研究

在详细分析了动态轮廓模型抗干扰性差、运算量大、不能逼近比较复杂轮廓、初始轮廓线设置复杂等问题原因的基础上,提出了轮廓线"有效逼近域"概念,进一步研究发现,这些问题都可以通过设置初始轮廓线到"有效逼近域"内,并有效地控制"有效逼近域"范围而得以解决。由于初始轮廓线只要求设置在"有效逼近域"内,因此符合基于小波变换多尺度边缘检测算法的特性,"有效逼近域"也可以通过控制GVF力场迭代次数来有效控制其大小。基于此,通过采用小波变换多尺度边缘检测获得不同分辨率边缘点,合理控制GVF力场迭代次数,提出初始轮廓线连接算法,成功地实现了初始轮廓线的有效设置。实验结果表明,能够准确地将初始轮廓线设置在"有效逼近域"内,并且"有效逼近域"的大小可以减小到真实轮廓左右5个像素以内,运算量有效的减小,抗干扰性也得到了很大的提高。

The Research of Contour Initialization Algorithm in GVF Snake Model

The concept of "effective area" is introduced in this paper based on detailed analysis of the problems of current active contour models: poor performance under interference, large computational cost, inability to converge to complex contours and difficulty of setting initial contour, etc. The above problems can be solved by setting the initial contour inside the so-called "effective area" and controlling the size of the "effective area" . Setting the initial contour inside the "effective area" is consistent with multi-scale edge detection methods based on wavelet transform. The size of "effective area" can be controlled by the number of GVF field iterations. Based on above discussion, a new contour initialization algorithm is implemented by connecting the initial points of different resolutions obtained using multi-scale edge detection methods based on wavelet transform and suitably controlling the number of GVF field iterations. Experiments show that the initial contour can be accurately set inside the "effective area" and the size of "effective area" can be reduced within 5 pixels from the true edges. The new contour initialization algorithm has also reduced computational cost and can perform reasonably well under interference.