Image Retrieval Based on Contourlet Texture and Scalable Color Descriptor

Wavelet transform cannot effectively catch the directional information of an image (only concerning horizontal, vertical and diagonal direction). The deficiency of directionality renders wavelet transform to underutilize the geometrical regularity of the image. Pointing to the problem, the paper substituted Contourlet transform for wavelet transform. Contourlet transform is the image 2D representation in the true sense of the terms, which has multi-resolution, localization and directional etc fine properties. Furthermore, there may be different and variable direction numbers in each scales. This dissertation analyzed the three statistics (mean, standard deviation and entropy) extracting from different subbands obtained by Contourlet transforming an image, and conducted the comparative analysis about the retrieval efficiency of Contourlet texture, wavelet texture and gray-level co-occurrence matrix. Finally, an image retrieval method was proposed by combining scalable color descriptor recommended by MPEG-7 with Contourlet texture. The experiment results show the effectiveness of the proposed method.