Study on the Correlation Properties of Wavelet Transform Coefficients and the Applications in a Neural Network-Based Hybrid Image Coding System

摘要

In the paper, the correlation properties of wavelet transform coefficients are investigated in details, including studies on the performance of wavelet transform as a de-correlator and the correlation along each prediction channel. Based on the study, a hybrid image coding algorithm is presented. With the proposed coding scheme, the original image is first processed by 1-D wavelet transform along one direction. A bank of neural networks will then be applied to the transform data to perform prediction along the other direction. Due to the high correlation existing in each prediction channel, the developed hybrid image coding approach can achieve high compression ratio compared with its counterparts.