针对传统匹配算法对旋转和扭曲图像匹配效果不佳的问题, 提出一种基于蛋白点区域SIFT(Sale Invariant Feature Transform)特征的凝胶图像间蛋白点匹配算法。首先, 提取蛋白点区域SIFT特征; 然后, 根据SIFT特征实现蛋白点粗匹配, 并采用RANSAC(Random Sample Consensu)方法剔除误匹配特征点; 最后, 通过计算粗匹配点集之间的TPS(Thin Plate Spline)变换关系, 采用几何相关法完成蛋白点间的精匹配。通过对国际凝胶图和Bio-Rad公司测试图等不同图源的凝胶图像进行蛋白点匹配实验, 结果表明, 该算法具有较高的匹配精度, 其匹配误差小于2.2% , 对旋转和扭曲图像同样具有良好的鲁棒性。

Since the traditional matching algorithm can hardly work in matching rotary and skewed gel images, a matching algorithm based on protein area SIFT feature is presented. Firstly, SIFT features are extracted in protein area. Then protein spots are matched coarsely by using SIFT features and mismatched features are removed by means of RANSAC. Finally, the correlation of TPS transformation is obtained based on the result of coarse matching. The precise matching in the rest of protein spots which is not matched in the process of coarse matching, is accomplished by the method of geometric correlation. The algorithm is applied to some source images including the international gel images and the gel images of Bio-Rad Company. Experimental results show that the proposed matching algorithm has higher matching accuracy, the matching error is less than 2.2%, and as well as better stability for rotary and skewed gel images.