A partial-duplicate image retrieval method using color-based SIFT
Alireza Pourreza, Kourosh Kiani · 2016
Many partial-duplicate image retrieval systems use the whole image for features extraction, while there is only a small duplicate region between the partial-duplicate images. On the other hand, many researchers consider the SIFT (Scale-Invariant Feature Transform) feature as an important descriptor in image retrieval systems, whereas it is independent of color and just describes the local gradient distribution, so, false-positive matches may occur in the matching task. To solve the problems, we propose a color-based SIFT matching method for partial-duplicate image retrieval that extracts the region with abundant visual content from image as a salient region. Then, keypoints are achieved from the region by applying SIFT, and color histogram is computed for each of them to improve the accuracy of image matching results. Our extensive experiments on IPDID and INSTRE datasets with 79% and 52% MAP measure respectively indicate the excellent performance of our method.