图象数据压缩的BP神经网络方法研究

BP Neural Network for Image Data Compression

重点研究BP人工神经网络用于图象数据压缩时 ,网络拓扑结构变化和算法修正对网络训练时间及重建图象质量的影响。仿真表明 :选择合适的网络结构 ,采用文中给出的快速网络训练算法 ,可明显加速网络收敛 ,且网络易避开学习误差的局部极小点 ,克服网络学习误差收敛刚性。

This paper gives the influence of the architecture and training algorithms of BP artificial neural network to the training time of network and image data reconstruction performance in image data compression. Simulation results declare that the training time of network is shortened markedly with suitable network architecture and the fast training algorithms given in this paper. Moreover, it is effective to add random values to synaptic weights and threshold to avoid premature saturation or local minimum in error surface. The data compression method of BP neural network gets a high compression ratio and good data reconstruction image performance.