Adaptive Modification of Transform Coefficients for Image Compression

For image compression purposes a general framework for modification of transform coefficients is proposed in this paper. To show the functionality of the method, we applied it to the contourlet transform. Unlike non-linear approximation (NLA) algorithms, the proposed algorithm causes the modification of the coefficients to be performed in a controlled manner. Different coefficients in different scales of the contourlet transforms have various roles in the reconstruction of the details of an image. The proposed algorithm adaptively modifies the coefficients based on the importance of the coefficient’s scale. We rank the scales and their coefficients so that the coefficients with higher impact scales are modified within smaller bounds. Larger modifications are applied to coefficients of lesser importance. All of the modifications are performed with the goal of minimizing the entropy of the coefficients. The implementation results show that our algorithm produces better numerical results and better visual qualities, especially for images with fine and regular textures, at low bit-rates.