A new cooling schedule in an annealed hopfield neural network for image vector quantization

Abstract

This paper shows an unsupervised parallel approach called the Annealed Hopfield Neural Network(AHNN) with a new cooling schedule for vector quantization in image compression. The main purpose is to combine the characteristics of neural networks and annealing strategy so that on-line learning and hardware implementation for vector quantization are feasible.Theidea is to cast a clustering problem as a minimization problem where the criterion for the optimum vector quantization is chosen as the minimization of the average distortion between training vectors. Although the simulated annealing method can yield the global minimum, it is very time-consuming with asymptotical iterations.In addition, to resolve the optimal problem using Hopfield or simulated annealing neural networks, designer must determine the weighting factors to combine the penalty terms. The quality of final result is very sensitive to these weighting factors, and feasible values for them are difficult to find. Using the AHNN to vector quantization, the need of finding weighting factors in the energy function formulated and based on a basic concept of the "within-class scatter matrix" principle can be eliminated and the rate of convergence is much faster than that of simulated annealing. The experimental results show that good and valid solutions can be obtained using the AHNN in image vector quantization. In addition, the convergent rate with different cooling schedule will be discussed.