A Method for Content-Based Image Retrieval with a Two-Stage Feature Matching

Content-based image retrieval is an active area of research where image content is used to guide the search of relevant images from a dataset. Given a query image, the images in the dataset are ranked in terms of their scores of similarity to this image based on their visual appearance. Many existing algorithms are based on either single feature or the fusion of multi-features with a one-step search method, which may lead to undesirable results due to the mismatch between low-level features and high-level semantics. To address this issue, we propose a two-stage sequential search algorithm where the color feature, represented by a color histogram in the HSV space, is used to form an image set containing images of similar color distributions to that of the query image, then a second stage of search is performed via the matching of feature points, in terms of discrete wavelet transform (DWT), and the scale invariant feature transform (SIFT) feature, extracted from a low-frequency subgraph. Experiments are performed on the ZuBuD dataset and UKBench dataset. Compared to some state-of-the-art algorithms, the proposed algorithm gives higher precision score.