直接基于灰度图象的多尺度角点检测方法

在计算机视觉领域中角点特征的提取具有相当重要的地位，因为角点包含了丰富的信息。本文提出了一种新的利用微分几何和小波变换的多尺度角点检测方法，它具有以下特点：一是无须提取边缘图象，而是直接在灰度图象上提取角点。二是由于利用了小波变换技术从多个尺度上进行角点的提取，所以它既较为有效的抑制了由于噪音等原因造成的诸多错误角点位置的出现，又能得到较为精确的有用角点位置。最后通过实验证明了本算法有效性。

A New Method of Gray Level Corner Detection Based on Multiscale Image

Coner detection plays an important role in the field of computer vision, because the corners include a large mount of useful information. In this paper, using differential geometry theory and wavelet transform, we propose a new method that can detect corners in multiscale images. Its special points as follow. 1. Needn't get edges tirst, we can detect corners in gray level directly, 2. To reduce the number of corners that introduce by noise and locate the important corners position accurately, we used wavelet transform to characterize corner in multiscale. The effectiveness of the new method is confirmed by the experiment in the end of this paper.