Polynomial- and Region-Fitting Schemes for Image Compression

The work''s main part is to implement a simple scheme to compress images through plane fitting. The scheme achieves better than 60:1 compression, while maintaining acceptable quality. Results are superior to JPEG at comparable performance. The scheme is multiplication-free, making it a perfect candidate for online and/or progressive compression. Blocking effects were reduced up to 0.85dB of PSNR through simple line fitting on block boundaries. The performance of the scheme is further improved by optimizing its predicted model parameters based on previously coded neighbouring blocks. It is found that less than 2 bits (on average) are enough to index the position of the candidate neighbour, making a 100:1 compression possible. The improvement in the compression ratio came at the expense of moderate to small quality degradations. A weber-based postprocessing is also introduced that improves the reconstructed image subjectively (rather than objectively). Higher order polynomial are also investigated, however region-based fitting seems to need further insight. As a by-product, weber law was extended to edge detection and quality assessment applications.

Autorentext

PhD in Electrical and Computer Engineering. 8 publications in signal/image processing. Supervised 14 graduation projects in various areas of image processing. Research interest in optimization, 2D and 3D fitting, and numerical analysis.