基于Gabor滤波器和聚类的图像配准算法

摘要

A novel registration algorithm is proposed based on Gabor filter of multi-scale and multiple direction and AP clustering. The objective image is decomposed into a sum of layers by Gabor kernel, and Harris corners are extracted in different directions on each layer. The local optimum of scale-space is searched within a window centering around Harris corners. The key points are described in terms of gradient in the characteristic region. The sets of feature point received from the two objective images are analyzed and the m:n coarse matching is realized. The corresponding points are matched in terms of Euclidean distance in each class. AP clustering can effectively eliminate many false matching between images with similar contents. Experimental results show that the proposed method accurately extract the steady feature points, and successfully remove mismatching between images with multiple similar contents.%提出一种基于多尺度、多方向Gabor滤波器提取图像局部不变特征并用AP聚类进行约束的配准算法。该方法首先利用Gabor尺度空间核函数对图像进行尺度空间分解，在每一层尺度图像的不同方向上提取Harris角点，在以Harris角点为中心的固定大小的搜索窗内搜索三维尺度空间的极值点作为局部特征点的位置和特征尺度；在特征子区域内用梯度描述特征点；将得到的两幅图像的特征点AP聚类分析，实现m：n的粗匹配，最终通过各类之间的欧式距离实现对应点的匹配，通过AP聚类可有效排除多相似内容的图像之间的误匹配。实验结果表明，该算法能够提取稳健的精确特征点，并且可以有效去除多相似内容图像带来的匹配误差，实现图像的配准。