基于曲率的透视不变性分割

PERSPECTIVE INVARIANCE SEGMENTATIONBASED ON CURVATURE

曲线的透视不变性分割在基于二次曲线的三维重建中非常重要。首先给出了连续曲线零曲率点和角点的提取法。在数字图像中曲线是不连续的，曲率的变化受噪声的干扰很大。在这种情况下为了检测出零曲率点和角点，对曲线采用了高斯滤波然后再提取零曲率点和角点。实验证明这种方法效果优于直接求曲率法。

Perspective invariance segmentation of curves is very important in quadric-based 3-D re- construction . First an extraction approach of zero curvature and corner points of curves in continuous case is given, In digital images, curvature of curves is not the same with the situation in continuous cases and very sensitive to noise or other disturbance in image. In order to get a robust detection ap- proach of zero curvature and corner points, Gaussian kernel is used to smooth curves firstly and to fur- ther extract zero curvature and corner points in curves . Our experimental results are promising com- pared with the direct detecting of curvature.