Statistical Dependences of Images Coefficients in Contourlet Domain: Analyzing and Modeling

In this paper, we propose a new scheme based on contourlet to compress the natural images with contours and oscillatory patterns. The contourlet transform, having the properties of multiscale, local, and multidirection expansion, is used as the non-linear image approximation method to decompose the images. With a detail study of the statistics of the contourlet coefficients of natural images, the individual behaviours and correlations of contourlet coefficients across scales, positions and directions are measured respectively. Results reveal that the contourlet coefficients have strong local dependencies and clustering when the coefficients are at low amplitude. Based on these findings, a new embedded block with significance sorting model is developed to present the transformed coefficients. Using the strategy of significance sorting, the proposed model can deal with both the significant coefficients in different subbands and the set of massive insignificant coefficients. Experimental results demonstrate that this approach is closed to the wavelet coder in terms of the PSNR metric, and visually superior to the wavelet coder for the images with detailed texture