[摘要]：We propose a new approach for detecting interest points using the support value of Gaussian function, which uses the support value to represent the salient features of the image. The support values image is computed by convolving the image with the support value filter deduced from the mapped least-squares support vector machines. The multiscale representation is built by successive smoothing of the support values image with a Gaussian kernel. Then, the normalized support value of Gaussian function is used to find the location of interest points and to select the points at which maxima are over scale. We compared our approach to the state of art of approaches using a standard data set. The experimental results show that the proposed approach performs better than other detectors for scenes under scale and blur changes, in terms of the repeatability score. The performance of the proposed approach is also confirmed by the image registrational results. Moreover, an extension of our method to airport recognition is presented. (C) 2010 Society of Photo-Optical Instrumentation Engineers. [DOI: 10.1117/1.3517746]
