Hybrid Predictive, Wavelet and Arithmetic (HPWA) Image Coding System

This paper presents a novel still image coding system that outperforms JPEG2000. In the proposed hybrid image coding system, the first stage involves performing predictive coding on the image using a dynamic ridge polynomial neural network as a nonlinear predictor which produces prediction errors. The prediction errors are fed to a novel Truly Adaptive Quantiser (TAQ)(second stage) which produces quantized prediction errors. In the third stage, the quantized prediction errors are fed to discrete wavelet transform which produces the transform coefficients and in the fourth stage an arithmetic coder is used for additional compression of the data and more effective coding of the transform coefficients. The system is completely reversible in order to re-produce the original image. The novelty in using nonlinear predictive coding as a front end for wavelet coding makes the data(quantized prediction errors) much more redundant and hence greatly improves the compression ratio of the system. Simulation results indicated that the proposed system shows improved performance in comparison to JPEG2000.