Content-based image retrieval by DTCWT feature

Abstract

This paper presents an multiscale image retrieval algorithm which uses a new class of image features as the image descriptors, from the DTCWT(Dual-Tree complex wavelet transform). It uses DTCWT to detect keypoints, make feature vectors which are invariant to image scaling, translation and rotation as similar as SIFT(Scale Invariant Feature Transform) feature, compare the similarity of the features with Euclidean distance, compute the similarity of the images with the statistic information of matching features. We do experiments in the Corel image database and compare with the AF-SIFT(Attention model based SIFT filtration). Experiments show that our algorithm is more appropriate for image retrieval.