Associative correction of compression artefacts with a self-organizing map classifying the image content

Summary form only given. A system for the correction of compression artefacts based on an associative memory is proposed. The system is further improved by incorporating a self-organizing map (SOM) to control the system parameters. The JPEG standard has been used for lossy image compression due to its inherent image capturing devices. Sub-images are employed because they contain limitations in variance and can be sorted into limited classes. In addition, the variability of these sub-images is often limited according to the major contents of the entire image. Internal coherences that exist in the sub-images of a particular class should be generally exploited for redundancy reduction. The missing information on the coherences within an image class can be recalled from an associative memory. The SOM classification of the sub-image content allows application of the correcting system to different types of images and blocks. SOM directly controls the parameter sets of the associative memory. Compression errors can be reduced depending on the image content and quality factor.