Fast Fractal Image Compression Parallel Algorithm And Implementation

Distributed computing can connect thousands of computers making use of network so as to form a virtual super computer, which can complete the problem of very large scale that a single computer can't finish. The real bottleneck of the information technology in new century is real-time processing and mass storage. And also the image processing and storage is the top priority. In recent years, fractal image compression has more advantages than traditional compression methods in the shape of compression effect, compression ratio and decoding time. It can meet the requirements of information age by improvements to a certain extent.This paper focuses on the research of algorithms based on a semi-norm of normalized block on the basis of the parallel network environment. The traditional image compression algorithm limits the practical application, because it costs too long time and the calculation amount is too much. For this problem, we make a detailed study of the algorithm based on reference [5]. Firstly, we analyze the possibility of parallel and apply this idea to this kind of encoding method. The design of the parallel algorithm is as follows:1. The tasks are divided according to the appropriate particle size on the principle of "Divide and Conquer". The particle size will affect the performance of parallel system directly.2. Applying the approach of parallel reading data, each processor reads file data independently according to their received numbers of tasks and encodes the data. The time of reading documents decreases with the increase of the number of processors.3. The tasks are assigned to the process according to the dynamic strategy. And all processed deal with data in a parallel way. The use of the dynamic strategy not only reduces the communication traffic of process but also reduces the overhead of management when running to achieve load balance.4. Build the appropriate particle size of the summary result. Firstly, the processors which participate in the operation collect the result independently. Then the processors transmit the result to the principal controller when then number achieves the particle size. It reduces the communication time of processors.In this paper, we discuss the optimal architecture pattern of parallel system in the improved algorithm and develop the dynamic task assignment scheme of "able person should do more work" to assure the load balance of system. We define and design a novel data structure of MPICH to improve the system efficiency. We implement the data transmission of high-performance by use of the packet transmission technology of MPICH. We provide the parallel algorithm based on the parallel environment. We build an experimental platform and do the experiment. The result shows that this algorithm can find out the best-matched block to an input range block in a relatively-small search neighborhood with a smaller time cost and also limit the search space. It also shows that this algorithm has a better performance in terms of encoding time and image quality.