Multiple kernel scale invariant feature transform and cross indexing for image search and retrieval

摘要

Image retrieval on large-scale image databases has attained more attention, in which mapping features into binary codes are showing great advancement. This paper proposes a new approach for image retrieval, Multiple Kernel SIFT (MKSIFT) that extracts the features from the pre-processed input image. It utilizes the steps of SIFT to extract the feature points. MKSIFT computes the keypoint descriptor with the introduction of exponential and tangential kernels, in which the weights assigned to the kernels are selected by Particle Swarm-Fractional Bacterial foraging optimization (PS-FBFO) algorithm. Moreover, it performs a cross-indexed image search by converting the feature points of MKSIFT into binary codes. The performance of MKSIFT+ Cross indexing is compared with that of SIFT, BSIFT, BSIFT+ Cross indexing, in which the proposed MKSIFT+ Cross indexing shows maximum performance. The experimental results evaluated the parameter precision, recall and F-measure provided maximum mean precision of 0.89793, recall of 0.8625, and F-measure of 0.87716.