A new method is proposed to extract features under affine transform rapidly. First maximally stable extremal regions(MSER) is detected in the image, then the regions are normalized and extracted by scale-invariant feature transform (SIFT). The experiments show that the method not only gives MSER mathematical description, but also is invariant to affine transforms. Additionally, the method has higher speed than SIFT because it only employ SIFT in maximally stable extremal regions.%为了快速稳定地进行仿射变换下的局部特征区域提取,提出了一种新的特征区域提取算法.首先检测图像最稳定极值区域(MSER),然后将这些区域归一化,接着对其使用尺度不变特征变换(SIFT)算子进行特征提取.试验结果表明,所提出的算法不仅赋予最稳定极值区域合理的数学描述,而且其抗仿射变换能力强于SIFT算子.另外,由于只是对部分图像计算SIFT算子,速度上也要快于传统的SIFT算法.
展开▼