中文摘要：

提出了一种基于遗传BP人工神经网络的COSM图像复原算法，利用BP神经网络的学习记忆和泛化能力，通过用一组COSM样本图像对网络进行训练，建立含有离焦模糊的模糊三维图像与清晰三维图像之间的非线性映射关系，然后利用训练好的BP神经网络对待复原的COSM图像进行复原处理，从而实现COSM图像复原．复原的三维图像无论在主观视觉还是定量分析上都取得了很好的效果．与传统的图像复原算法不同，该算法免去了解卷积等复杂的运算，不存在病态问题，可广泛应用于模糊图像的复原中并且效果较好．

英文摘要：

A new method is proposed for image restoration of COSM based on the genetic BP neural network. The nonlinear mapping relationship between the 3D blurring images with defocusing message and 3D clear images are established by training the BP neural network which has the ability of learning, remembrance and generalizing with a group of COSM images whose blur style are known. Then 3D image which needs restoring could be restored by the trained neural network.. As a result the restoration of 3D image is achieved. Extensive tests demonstrate that this method has a satisfying restoration effect both in visual impression and quantitative analysis. The method can avoid deconvolution and ill problems which are different to the conventional image restoration method. It could be widely used to restore blurred image and has a good effect.