Nonlinear local contourlet energy pattern for image retrieval applications

Local patterns are effective in different machine vision problems such as pedestrian identification, lane categorization, face recognition, retrieving required image etc., Various local patterns are introduced by the researchers in order improve the efficiency, however these local pattern operates on a fixed pixels that are predetermined and are same for all the images. The features thus extracted from these predetermined pixels are limited. In this paper a novel technique called nonlinear local contourlet energy pattern (NLCEP) is introduced which extracts the local pattern from the pixels that are selected dynamically in run time which will vary with the images. Also to improve the feature robustness the features are extracted in Contourlet domain instead of the spatial domain. With this approach the dominant image features like lines/curves are better represented by the NLCEP and its features are effectively used in image retrieval system. The performance of this method is validated by doing different experiments with the standard available databases (viz Corel 1K, Corel 10K and Brodatz). The test results with different experiments shows that the proposed approach provides better performance for image retrieval applications.