Ornament Image Retrieval Using Multimodal Fusion

Search-by-example, i.e. finding images that are similar to a query image, is an indispensable function for various modern image search engines. The applications of such systems are manifold. The primary application of search-by-example is in recommending fashion materials based on user interests. There are various challenges in this area of research such as a large volume of the product database, similar visual appearances, and a large variety of products. The problem becomes more difficult to solve when the product is complex in design such as ornaments. In this paper, we have proposed a fusion-based retrieval model. The method uses weighted average of multiple similarity measures. We have used four different methods namely hash-based, histogram-based, deep feature comparison, and feature cross correlation to find the similarity. A dataset of ornaments (golden earrings) has been prepared and made available to the research community. We achieve 81% top-1 and 89% top-5 accuracy using the proposed method. The dataset and the code is available publicly in https://github.com/skarifahmed/RingFIR .