Digital Image Compression Method Study Based On Artificial Neural Networks

Human get much information from the image media, and image has become a most important vector in human life at exchanging information. What's more, digital information processing is an important symbol in the information society.As the image information digitized, there are many biggest problem, such as the massive data storage and transmission. Therefore, the data compression technology, especially digital image data compression technology is a key to the solution. Reduce the redundant data of the image to achieve image compression purposes, and make full use of the human eye's visual identity and image of the statistical features to reduce image information redundancy, then to do images compessing.For these reasons, a variety of digital image compression technology came into being. Which, based on artificial neural networks in digital image compression technology has become a hot research topic.Artificial neural networks background to the study which began in the late 19th and early 20th century. It originated from physics, psychology and neurophysiology of interdisciplinary research. Its network structure is varied. The networks generally can be divided into two categories: feedforward networks and feedback networks. Feedforward networks contain BP network, MADLINE network, multilayer perceptron network (MLPN), radial basis function network (RBFN), etc.; while the feedback network typically includes Hopfield network, Boltamann machine, ART networks.One for digital image compression neural network generally: Based on error back propagation MLP - BP neural network, self-organizing feature map - SOMF network, and the Propagation Neural Network - CPN net.This paper, we mainly based on BP neural network, SOMF network and CPN network to do digital image processing compression application and to introduce digital image compression of experimental methods and results. We do image compression through four steps such as the image data preprocessing, network building and training, simulation and image reconstruction to. And then use signal to noise ratio and peak signal to noise ratio performance function to evaluate image compression quality.The experiment shows that artificial neural networks, although only a human brain, a simple modeling and abstraction, but they ar much similar to the human brain whose information processing capacity, therefore the networks have strong data compression capability. In addition, the neural network will also promote the image coding to the intelligence and knowledge transformation. And the strong fault tolerance and associative memory function of the neural network contribute to digital image processing.