Iterative projection algorithms for reconstructing visually pleasing images from block discrete cosine transform (BDCT) compressed image data are presented. Two algorithms are proposed. The first is based on the theory of projections onto convex sets (POCS). The second is motivated by the theory of POCS. In order for the POCS approach to be effective, the convex sets must be as pertinent as possible to the required objective. For the deblocking of the DCT compressed images two types of constraint sets capture most of the information pertinent to this problem: the set that contains the information of the transmitted DCT coefficients and the sets that express the between-block smoothness properties of the desired image. The proposed iterative reconstruction algorithms have been tested thoroughly on different DCT-based compression approaches for a number of 256*256 and 512*512 images. The experimental results demonstrate that the proposed algorithms yield superior images to those obtained by direct reconstruction from the compressed data only.
展开▼
