Recovery Methods for Postprocessing of Compressed Images

This chapter presents recovery-based techniques for compressed image postprocessing—that is, projections onto convex sets (POCS) and the maximum a posteriori (MAP) methodologies. This chapter also presents the basic theory of the POCS methodology and the application of POCS to the compressed image postprocessing problem. The application of the MAP methodology is also presented in the chapter. The objective of postprocessing is to improve the quality of the images produced in the decoder of a lossy image compression system. Such systems produce high compression ratios; however, to do so they also discard information, which is deemed not important, and thus introduce distortion to the original image. The distortions produced by lossy compression algorithms can be categorized into two classes. First, all lossy compression algorithms produce what is called “ringing” artifacts. Around sharp intensity transitions in the image, these oscillations have been classified to be of the Gibbs type. These artifacts appear at high compression ratios in all transform-based codecs because of the low-pass nature of such systems. Second, the classic JPEG compression algorithm at high compression ratios produces what is called the “blocking” artifact. This artifact originates from the independent quantization of the block discrete cosine transform coefficients, which is used in the classic JPEG algorithm.