Improving Texture Image Retrieval with Redundant Contourlet Features

In this paper, we develop RCT-Plus, a new variant of the contourlet transform in which augmented directional selectivity, redundancy and Gaussian filtering are meant to enhance texture characterization and feature extraction for content-based image retrieval (CBIR). Two modeling methodologies namely, generalized Gaussian distribution and energy computation, are applied on contourlet coefficients. The features that are derived from the estimated model parameters may rely to both directional and approximation sub-bands, providing thereby various CBIR frameworks. Experimental results show that RCT-Plus achieves superior retrieval performance compared to other methods using standard contourlets. Therefore, the contribution of RCT-Plus is significant to the relevance of the features extracted from textures.