Um modelo de compressão de imagens digitais baseado em quantização vetorial e transformações afins

Abstract

The fractal compression is an emerging digital image coding technique which explores the self-similarity present in digital images, showing good fidelity between the original image and the reconstructed image, achieving high compression rates. However it has some weaknesses and because it demands a considerable computational complexity commonly is used the assistance of one or more techniques to meet those needs. This research presents a model which combines fractal coding and vector quantization (VQ). In addition, the time spent in choosing geometric transformation (isometry), which is an important step of fractal coding, is minimized by being made in the frequency domain by the DCT (Discrete Cosine Transform) inner product. The LBG (Linde_Buzo_Gray) algorithm is used to designate a generic codebook that replaces the traditional domain-pool of a fractal coder. The result is a hybrid coder with better performance than the pure fractal coders that preserve the visual quality of the reconstructed image and reaches high compression rates.