Study On Fractal Image Coding Based On Partheno-Genetic Algorithm

Image is the most important carriers for communication in our life, and it's a medium containing the largest information. It is well known that the data quantity of digital image is so large that it is absolutely necessarily to be compressed before it can be stored or transformed efficiently. So image encoding has become one of the key technologies on which the digital society depends.Fractal image coding is a very promising image compression technique. It has been receiving more attention in research over the past decade because of its novelty,high compression rate,resolution independence,but too long encoding time is still the main obstacle that prevent it been put into practice. The expensive computational cost come from process of finding the best matched domain block in a large domain pool for a given range block, which is the optimal matching problem. In fact, it can be considered as a multimodal function optimization problem.Genetic algorithm is a heuristic random searching strategy based on the principle of nature selecting and genetics, this dissertation focus on solving the optimal matching problem in image fractal coding with genetic algorithm, the main contributions are as follows:(1) It analyses the characteristic of the optimal matching problem from the point of view of multimodal function optimization, then presents a real-coded Partheno-Genetic algorithm based on the local self-similarity of nature image.(2) It proposes a adaptive species deviation method according to the individuals cumulate marginal distribution density curve, then on the basis of which presents a strategy based on penalty function and taboo-search to prevent premature convergence.(3) It carries out numerical simulation of basic partitioned fractal coding method and the proposed method.Experiment result suggests that at the costs of almost unnoticeable degeneration of decoding image quality, the proposed algorithm can greatly speed-up the coding process,the comparing experiment result also shows that the performance of our algorithm similar to the binary-coded Genetic algorithm with crossover operator.