Estimation of edge parameters and image blur from local derivatives

Fractal image coding has been used successfully to encode digital grey level images. Especially at very low bitrates fractal coders perform better than cosine-transform-based JPEG coders. A block-based fractal image coder is able to exploit the redundancy of grey level images by describing image blocks through contractively transformed blocks of the same image. Previous fractal coders used affine linear transformations in combination with 1st order luminance transformations that change the brightness and scale the luminance values of image blocks. We propose an extension to high order luminance transformations that operate in the frequency domain. With this transformation and an adaptive coding scheme a better approximation of image blocks can be achieved. Bitrate reductions are higher than those achieved by "spatial-domain" fractal coding schemes. An additional effect of this new transformation is a better convergence at the decoder.