Chaotic Neural Network-based Image Compression Technology

Digital image processing techniques have been used more and more widely in many fields such as multimedia, Internet,television and fax, etc. Image compression is one of the most important key techniques in digital image processing. Traditional compression methods include prediction coding, transform coding and vector quantization(VQ). In the last twenty years, modern techniques based on neural networks, fractal theory and wavelet transform have been successfully used for image compression.Chaotic Neural Networks(NN) is a new research field which has been developed in these years. It has been studied more and more because of its complex dynamic characteristics. Normal NN has characteristics of gradient descent, while Chaotic NN is different. NN with chaotic characteristics not only has more abundant and far equilibrium point dynamic characteristics but also has attractors. Because of its complicated dynamic characteristics chaotic NN becomes a widely used technology in information processing and optimization.This dissertation focuses on the applications of chaotic neural networks in static image compression. It includes two main parts. First, two types of chaotic neural networks based on Logistic mapping are researched. They areⅠandⅡfeed-forward chaotic neural network. The application and key technology of them in image compression are discussed. By many simulation experiments, compressive property of two models is compared with earlier methods. Second, a novel chaotic NN is researched by analyzing the working principle of existing chaotic neural networks. Similar to the structure of Hopfield NN, the neural networks have transient chaos response which has more abundant dynamic features and stronger global search capability. Combining chaotic dynamics and convergence mechanics, the neural networks is changed gradually form chaotic NN to Hopfield NN as to control the chaos, and the initial value near global optimal solution is also provided at the same time. Through the optimization function of the model and a competitive model which resolve the design of codebooks, the energy function and dynamics equation of neurons are proposed. An algorithm of codebooks design is also presented, so as to the model is successfully used in image compression of vector quantization.The simulation results show that three algorithms of image compression based on chaotic NN have obvious advantages in training speed, compression quality, robustness etc.