Using Tsallis entropy into a Bayesian network for CBIR

This paper presents a Bayesian network model for content-based image retrieval (CBIR). In the explanation and test of this work, only two images features (semantic evidences) are involved: color and shape (from gradients of directions). However, one of the main advantages of the proposed strategy is its easy extension to several evidences. Considering the precision with which the images are retrieved, to highlight the evidences that generate the best results, we have introduced the use of nonextensive entropy. This concept extends the Shannon´s classic theory of entropy for information systems. Experimental results show that may be a link between the parameters of the Tsalli´s nonextensive entropy and the precision with which the images are retrieved from the database. In some cases, we have obtained up to 30% in terms of average precision.