基于有向面积的角点分类算法

A Corner Classification Algorithm Based on Directed Areas

在3维场景重构、运动估计、机器视觉等领域,不仅希望能检测出图像中的角点,而且还希望获得角点附近更多的信息,即能对角点进行进一步分类。为了能对图像中检测出的角点进行分类,提出了一种基于有向面积的角点分类方法,该方法首先采用基于协方差矩阵特征向量的小波变换角点检测原理检测出图像边缘上的角点;然后根据角点两侧的边缘信息定义了6种类型的角点;最后通过计算角点附近边缘上顺序排列的3个有向面积,实现对角点的分类。实验表明,基于有向面积的角点分类,具有较高的准确性。

In the domain of 3D scene reconstruction, motion estimate and machine vision, we need to detect the corners in a image, as well as obtain more information around corners, namely we could classify the corner further. We proposed a corner classification method based on directed area. First, we applied corner detection principle of wavelet transformation, based on the eigenvectors of covariance matrices, to detect the corner on the edge of images. Secondly, we defined six types of corners based on the edge information of both sides. Finally, corners can be classified by computing three directed areas arranged in order on the edge of the corners. The experiment indicated that the method of classifying corners has higher accuracy.