Research On Image Compression Method Based Upon Improved BP Network

Some current methods of image compression technology are described briefly in this dissertation. The superiority and possibility of using neural network to compress image are explained too. The principle of the image compression based on BP network is also mainly analyzed in detail. In order to enhance the training speed of the BP algorithm, the quality of image compression and compression ratio, two improved methods are brought forward:Firstly, the BP network is optimized by the particle swarm optimization (PSO) algorithm. Aiming at the shortcomings of the normal PSO algorithm, an improved algorithm is proposed. By the experimental confirmation, the convergence ratio and speed of the improved PSO algorithm are all increased. Then, considering the merits of the PSO and BP algorithms, the training process of BP network is divided into two steps. The experiments indicate that the convergence speed and precision of BP networks using the improved algorithm are enhanced greatly. The compression ratio and reconstruction quality of images are improved distinctly.Secondly, the local learning rate adaptation algorithm——SA algorithm is improved. Many kinds of the global learning rate adaptation and local learning rate adaptation algorithms are introduced in this dissertation. And their merits and shortcomings are also presented. Then an improved SA algorithm is proposed. Through the test of the curve fitting, the convergence precision, convergence speed and stability of the improved SA algorithm are enhanced clearly. The image compression experiments show that the reconstruction quality and compression ratio are improved obviously.