基于 MSER 与 SIFT 融合的多通道图像匹配算法研究

摘要

针对传统最稳定极值区域算法,只能从灰度通道进行不变特征提取而不能充分利用彩色图像信息的缺点,以及 MSER(maximally stable extremal regions)算法与 SIFT(scale invariant feature transform)算法本身存在的局限性,提出一种融合 MSER 与 SIFT 的多通道图像匹配算法。首先利用 HSI(hue-saturation-insensity)颜色空间,同时从三通道进行 MSER 特征提取,对得到的特征进行简单筛选,将 S 和 I 通道得到的 MSERs 经椭圆拟合与归一化处理,并在 H 通道得到的区域中进行 SIFT 特征提取,而后对所有特征利用融合了 C-均值聚类法的 SIFT 描述子进行描述,并用高维数据的可伸缩最近邻算法(fast library for approximate nearest neighbors,FLANN)连接同名点,利用 PROSAC(progressive sample consensus)算法去除错误点,最终实现图像之间的匹配。仿真实验证明,该方法能够解决传统 MSER 存在的问题,并且在提取的特征质量以及处理速度上更具优势。%To overcome the disadvantages in the maximally stable extremal regions (MSER) algorithm,which can only detect the invariant feature in the gray channel while wasting the color image information, an extraction algorithm of integrated MSER and SIFT features in multichannel images is proved.First,MSER features are extracted from the three channels of HSI to make full use of the information of the color image.Then the process the MSERs obtained from the S and I channels elliptically fitted and normalized,and the SIFT features in the area detected from the H channel at the same time are extracted.After that,all of the features by the SIFT descriptor which combined the C-means clustering algorithm are described.And then the homonymy points by FLANN algorithm contacted.Lastly,PROSAC algorithm is used to realize the matching between images.The method is proved by simulation and it can solve the problem existing in the traditional MSER algorithm and achieve superior results in the feature quality and processing rate.