图像特征匹配算法评价框架

Image feature matching technology has a wide application in the image mosaic, digital photogrammetry and other field. Image feature matching algorithm develops rapidly for the diversification of the demand.Differentimage feature matching algorithm has different characteristics, so it is necessary to carry out comprehensive comparison of the existing algorithms, and make the optimal choice according to the different application. Based on the summarization of feature matching evaluation index used in comparing matching algorithms previously, a comparative framework is proposed. Inside the framework, the single index of image feature matching algorithm, with the use of the standard image data, is acquired and normalized. The comprehensive index value is gained by the weighted single index value and the index weight is set according to the different application background. Therefore, this framework can provide reference for selecting the algorithm based on the diversified application. With the utilization of the standard data sets provided by Mikolajczyk and Schmid, quantitative analysis of SIFT, SURF, BRISK, ORB and FREAK, in terms of time efficiency and stability of image transformation of rotation, blur, illumination, scale and visual angle, is carried out, and thus the single index of each algorithm is achieved respectively. Moreover, the rationality of the framework has been illustrated with the example of the image matching of the vehicle-borne mobile measurement system.