基于连通区域的仿射不变区域提取方法

摘要

To resolve the problems of high computation and low robustness of existing affine invariant feature extraction methods, a new affine invariant region extraction method based on connected region is proposed. The input gray image is filtered with an anisotropic gauss filter and then histogram equalization is performed. Connected regions with the same pixel values are found. Adjacent connected regions whose gray values' difference is less than delta are merged. The ultimate merging result that satisfies the demand is concerned as the image's local affine invariant regions. Experiments show that this method is very effective, fast and highly robust.%针对现有仿射不变特征提取方法存在计算量大、鲁棒性差的问题,提出一种基于连通区域的仿射不变区域提取方法.对输入的灰度图像进行各向异性高斯滤波,并对滤波后的图像进行灰度直方图均衡化.找到图像中灰度值相同点所组成的连通区域,将灰度值差小于delta的相邻连通区域进行合并,把满足条件的最后一次合并结果作为图像的局部仿射不变区域.实验证明,该方法提取效果好、速度快且鲁棒性强.