Dictionary Design for Block-Based Intra-image Compression

A very crucial question in sparse approximation-based image compression is the choice of the dictionary. An effective dictionary can lead to excellent representation of an image with least number of dictionary atoms and lead to excellent image compression. The performance of block-based intra-image compression scheme depends on how well the residuals obtained from intra-prediction are encoded. Transform-based intra-compression method uses fixed DCT dictionary for encoding of prediction residuals. However, DCT dictionary is not suited for efficient encoding of prediction residuals with complex and non-periodic characteristics. This paper presents an algorithm to design an over-complete residual dictionary suitable for encoding prediction residuals. Simulation results demonstrate that the proposed residual dictionary yields superior performance as compared to other standard methods.