中文摘要：

针对基于深度学习的图像去模糊方法无法有效地保留高频纹理信息,易产生振铃效应,且时间复杂度较高的问题,提出基于卷积神经网络(CNN)的图像去模糊方法.该方法设计了一种高频信号保持且可快速去模糊的快速CNN模型(FCNN).在此基础上,首先对高频图像进行傅里叶域上的预处理,通过实施傅里叶域去模糊的预处理得到一个初始的清晰图像;然后将该初始图像小块作为输入,相应的真实清晰图像小块作为标签训练FCNN,得到从模糊图像到潜在清晰图像的映射函数,实现基于该训练网络的去模糊.定性和定量实验结果表明,文中方法利用CNN参数共享的特点,减少了网络训练过程中大量的学习参数;相对前人基于深度学习的去模糊方法,该方法对模糊图像在保持图像纹理细节恢复的同时使计算复杂度得到显著降低.

英文摘要：

Existing deep learning based image deblurring methods are weak in preserving the high frequencytexture details.They may also be time-consuming and cause severe ringing effects in the restored result.Thispaper proposes a novel approach based on the convolutional neural networks(CNN)to overcome these limits.A CNN model called fast CNN(FCNN)is introduced to deblur the image quickly while keeping the highfrequency details.With FCNN,the method first deblurs the image in the Fourier domain with regularizationsto obtain a pre-processed latent image.Then,the block of the pre-processed results and its correspondingclean blocks are served as input and label to train the FCNN,so that the mapping function from blurringimage to potential clean image is obtained.Consequently,a clear image can be obtained by the trainedFCNN.Qualitative and quantitative experiments demonstrate that the proposed method adopts the parameter-sharing property of CNN effectively and reduces the number of training parameters significantly.It alsogreatly reduces the computational complexity in comparison with the existing deep learning based algorithmswhile keeps the image details well.