基于各向异性高斯方向导数滤波器的角点检测

Corner detector via anisotropic Gaussian directional derivatives filter

为了改进噪声鲁棒性和定位准确性,利用各向异性高斯方向导数滤波器,提出多方向角点检测算法.该算法利用一组各向异性高斯方向导数滤波器对输入图像进行卷积处理得到各个方向的滤波器响应.对于每个像素点,利用它与周围邻近像素点的滤波器响应的相关信息构造局部自相关矩阵,然后根据自相关矩阵归一化特征值及像素点处各方向滤波器响应,作阈值处理和非极大值抑制处理判定像素点是否为角点.实验结果表明,在无噪声和噪声的条件下,提出的检测方法与各向同性高斯核函数的Harris算法相比,配准角点数均提高6.0%左右,具有更好的检测性能.

A multi-directional corner detector using the anisotropic Gaussian directional derivative filters is proposed to improve the noise-robustness and the localization accuracy .T he proposed algorithm first uses a set of anisotropic Gaussian directional derivative filters to smooth the input image .For each pixel , its filter responses and those of its neighborhood are used to construct the auto-correlation matrix ,then a corner measure combing the normalized eigenvalues and the responses at the pixel is calculated and threshold and non-maximum suppression are used to decide corners .The experiments show that ,com-pared with Harris algorithms ,the correct detected corner numbers by the proposed algorithm increase by about 6.0% for both noiseless and noise images .