A Learning-by-Example Method for Reducing BDCT Compression Artifacts in High-Contrast Images
A Learning-by-Example Method for Reducing BDCT Compression Artifacts in High-Contrast Images
Guangyu Wang

The quantization procedure of block-based discrete cosine transform (BDCT) compression (such as JPEG) introduces annoying visual artifacts. In this thesis, we propose a novel learning-by-example method to reduce BDCT compression artifacts in high-contrast images (images with large smooth color areas and strong edges/outlines), for example cartoon images. Our main focus is on the removal of ringing artifacts that is seldom addressed by existing methods. In the proposed method, the contaminated image is modeled as a Markov random field (MRF). We 'learn' the behavior of contamination by extracting massive number of artifact patterns from a training set, and organizing them using tree-structured vector quantization (TSVQ). Instead of post-filtering the input contaminated image, we synthesize an artifact-reduced image. Utilizing the proposed method, we show that substantial improvement (both statistical and visual) is achieved. Our method is non-iterative and hence it can remove artifacts within a very short period of time.

Related publication:

  • " Deringing Cartoons by Image Analogies",
    G. Wang, T. T. Wong and P. A. Heng,
    ACM Transactions on Graphics, Vol. 25, No. 4, October 2006, pp. 1360-1379.

  • "A Training-Based Method for Reducing Ringing Artifact in BDCT-Encoded Images",
    G. Wang, T. T. Wong and P. A. Heng,
    in Proceedings of the 7th Eurographics Workshop on Multimedia (EGMM 2004), Nanjing, China, October 2004, pp. 105-113.