一种旋转和光照不变的加速鲁棒特征算法  

Rotation and illumination invariant modified SURF algorithm

作　　者：徐志军[1] 耿则勋[1,2] 卢兰鑫 乔玉庆 沈忱[3] 

机构地区：[1]信息工程大学,郑州450002 [2]许昌学院城乡规划与园林学院,河南许昌461000 [3]陆军航空兵学院飞行模拟训练系,北京101123

出　　处：《计算机应用》2016年第A01期113-116,121,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(11373043)

摘　　要：针对加速鲁棒特征(SURF)算法对光照变化敏感、主方向误差影响匹配性能的问题,提出了一种改进的SURF算法。首先分析了特征邻域划分方式在旋转变换下对描述符匹配性能的影响,然后提出了采用圆形特征邻域构建描述符的方法,以消除主方向计算误差带来的影响。此外,利用自适应Gamma校正来克服光照变化带来的影响。在特征邻域内计算灰度均值,并将此与灰度级的一半的比值作为Gamma值,对特征邻域作Gamma校正,然后计算特征描述符。最后进行了匹配实验,与SURF算法进行了比较。实验结果表明,改进的算法具有光照不变性和旋转不变性,与SURF算法相比,在AMC580影像上正确匹配数增加了77.4%,匹配准确率提高了11.6%。The Speeded Up Robust Feature（ SURF） algorithm is sensitive to illumination changing and the dominant orientation error influence its matching performance.To solve the problems,an improved SURF algorithm was proposed.First,the influences of rotation on the descriptor matching performance of the feature region partition method were analyzed.Then,a method to construct SURF descriptors with circle feature region was proposed,for eliminating the influence of the dominant orientation computation error.Moreover,a self-adaptive Gamma correction was used to eliminate the influence of illumination changing.The Gamma values were computed by the ratio of the average gray value of feature region and half of the gray level,then a Gamma correction was done on feature region,and the descriptors were computed.Finally,the comparison experiments with SURF were performed.The experiment results show that the proposed algorithm is rotation and illumination invariant;for AMC580 images,the correct matching number raised by 77.4% and the correct matching rate increased by11.6% as comparing with those of original SURF.

关 键 词：加速鲁棒特征 旋转不变 光照不变 GAMMA校正 影像匹配 特征邻域 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]