提供了一种基于神经网络的DPCM的编码方法。它通过基于图象实例的训练，可以处理一些线性预测方法所不能处理的高阶特征以及非线性相关性。并由此方法设计了一种典型的预测器—2D-NNADPCM预测器，且提出根据不同图象的信息和要求进行人为分块，以达到更好的编码效果。

In this paper, a neural network approach to DPCM(differential pulse code modulation) design for encoding of images is presented. This approach can deal with high-order statistics and nonlinear correlation which the linear method can not. Based on this method, this paper provides a predictor—2DNNADPCM, and presents an approach that divides an image into different sub-block according to different messages and requirements in order to achieve better results.