Feature Space Optimization for Content-Based Image Retrieval

Substantial benets can be gained from eective Relevance Feedback techniques in content-based image retrieval. How-ever, existing techniques are limited due to computational cost and/or by being restricted to linear transformations on the data. In this study we analyze the role of non-linear transformations in relevance feedback. We present two promising Relevance Feedback methods based on Ge-netic Algorithms used to enhance the performance on the task of image retrieval according to the user's interests. The rst method adjusts the dissimilarity function by us-ing weighting functions while the second method redenes the features space by means of linear and nonlinear trans-formation functions. Experimental results on real data sets demonstrate that our methods are eective and the re-sults show that the transformation approach outperforms the weighting approach, achieving a precision gain of up to 70%. Our results indicate that nonlinear transformations have a great potential in capturing the user's interests in image retrieval and should be further analyzed employing other learning/optimization mechanisms1