利用并行神经网络实现分形图像编码时 ,提出一种限定父块选取的编码方法 .该方法保留了结构简单、计算量小等特点 ,获得的分形码流不仅可从任意图像开始 ,通过多次迭代重建图像 ,而且还可以从均匀灰阶图像 (全黑图像除外 )出发 ,迭代一次获得精确的图像边缘 .模拟结果表明该方法是有效的 . When using parallel neural network to realize fractal image coding, a new encoding method is proposed to limit the selection of parent block, which preserves the characteristics of simple structure and small amount of computation. The fractal code stream can not only start from any image, Reconstruct the image, but also get an accurate image edge from the uniform grayscale image except the all-black image, and the simulation results show that this method is effective.