主要讨论了经过小波变换后较难压缩的低频子带系数特性，并据此利用基于BP算法的多层前向神经网络高度非线性映射能力，构造了一个非线性预测器，有效地去除低频子带系数间的冗余度，在较高的保真条件下获得了罗满意的压缩比。

After discussing the second level low frequencies part of wavelet transform, which is rather difficult to be compressed, we can use the nonlinear mapping capacity of multi layer free forword neural network on BP algorithm. A scheme of image coding based on nonlinear predictor is put forward. Redundance among the wavelet coefficients can be ridded effectively with the help of this method. On the condition of relatively low distortion degree , the rather satisfactory compress rate can be achieved.