A novel content based image retrieval approach by fusion of short term learning methods

Relevance feedback is a powerful tool in Content based image retrieval (CBIR) systems that bridges the semantic gap and improves the performance of the system by interacting with user. In this paper, we merge the retrieval results of two short term learning (STL) algorithms using Borda count fusion method to improve the accuracy of the system. The proposed fusion method uses the advantages of individual STL algorithms by combining the ranked lists. To evaluate the proposed method, we implement a CBIR system in which each session consists of four rounds of relevance feedback and Corel data set with 10000 color images from 82 different semantic groups are used. The experimental results on 100 test images revealed that the combination method significantly outperforms the individual STL methods in terms of precision.