An image retrieval model providing semantics and visual features based query for users

Image retrieval is the hot point of researchers in many domains. Traditional text-based query methods use caption and keywords to annotate and retrieval image database, which often consumes a mass of human labor. Content based image retrieval methods use low-level features such as, color, shape and texture to search images, which can't provide retrieval on semantic level for users. In this paper, we propose a novel image retrieval model that provides users with both semantics based query and visual features based query. Our approach has several advantages. First, it integrates visual features and semantics seamlessly. Second, it uses some effective techniques such as image classification, relevance feedback to bridge the gap between visual features and semantics. Third, it proposes several ways to obtain the semantic information of the image, which reduces manual labor and reduces the "subjectivity" of semantics by human. Fourth, it can update semantics of the image by human's intervention, which makes the image retrieval more flexible. We have implemented an image retrieval system ImageSearch based on our proposed image retrieval approach. Experiments on an image database containing 22000 show that our scheme can achieve high efficiency.