Weighted semantic fusion of text and content for image retrieval

The performance of image retrieval (IR) systems improves by reducing the semantic gap between the low-level features and the high-level concepts. Research results in the recent years show that combining the two modalities (text based and content based) even with simple fusion strategies alleviates the image retrieval results and also reduces the semantic gap. In this paper, we propose a new approach called weighted semantic similarity, which assesses the semantics between the query image and textual query provided by the user as an input to the system. The similarity between the keywords has been measured the using WordNet. For content matching, color feature is extracted and is represented using Fuzzy Color Histogram (FCH). The two modalities are fused together using reordering technique to improve the retrieval results. The proposed approach shows that the semantics learned at an early stage not only reduces the semantic gap but also decreases the computation time largely. The Mean Average precision (MAP) of 0.4311 is achieved using the proposed approach.