Abstract
The Scale-Invariant Feature Transform (SIFT), which is invariant to image scaling and rotation, and partially invariant to change in illumination and 3D camera viewpoint, is proved to be one of the best local feature descriptors. However, the number of SIFT features for one image often varies from tens to thousands, and matching such number size of SIFT keypoints between two arbitrary images can bring vast computational costs. We proposed a method to reduce original SIFT features by filtering out those SIFT features which can truly represent category properties-Category Specific SIFT Descriptor (CSSD) based on clustering and statistics and then using them to reduce each image SIFT features. On the other hand, SIFT features are mainly designed for gray image rather than color image applications. Whereas the color information provides important contents as well, and it is rotation and scale invariant which never violates SIFT properties. Thereby we presented a content-based image retrieval system in which the reduced SIFT features are combined with color information via a loosely-coupled style. Comparing to some existing color-SIFT descriptors, this enables the system to easily control the weight between SIFT features and color information and never enlarge the number of SIFT features. Finally, extensive experiments showed that the proposed approach outperforms the original SIFT and the optimal color features from four popular approaches for combination and the optimal combinational weights for 10 image categories based on Corel image database are obtained.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call