基于边缘匹配和多尺度小波变换的图像校准

Algorithm of image registration based on edge matching and multi-scale wavelet transformation

提出了一种新颖的图像校准算法 .先将边缘方向曲线链码化 ,选出相对于基准点位置相似且局部链码匹配的对应边缘曲线段 .然后采用多尺度小波变换提取对应边缘曲线段上的真实角点 ,即得到两幅图的对应特征点集 .仿真试验证明 ,我们的算法可以在实时系统中 ,以较小复杂度 ,达到较好的性能 .

The kernel of Image Registration algorithm was the matching of edge′s relative Chain Code and Corner Detection method based on Multi-scale Wavelet Transformation. This algorithm constructed chain codes of the edge's direction curve. If an edge pair was locally similar in chain code and had little difference in their position (relative to a fiducical pair between the two images), they were matching edge pair. The true corners in these matching edges pairs were extracted according to results of Multi-Scale Wavelet Transformation. By means of that, matching corner pairs were gotten. Experiments showed that the algorithm performances well in real time system with low cost of time and space, compared either with the algorithm based on mutual Information or the algorithm based on Module Matching.