基于深度卷积神经网络的运动模糊去除算法

Motion deblurring algorithm based on deep convolutional neural network

为了提高图像去模糊的复原效果和处理速度,提出了基于深度卷积神经网络的运动模糊去除算法。以多尺度图像为依据,构建了基于自动编码器的网络模型。在扩大感受野方面,采用空洞卷积模块提取图像多尺度特征信息,采用残差模块拓宽网络深度,以解决训练过程中图像细节丢失的问题,实现了图像的端到端运动模糊去除任务。在GOPRO数据集和真实测试集上的实验结果表明,该文算法在参数量仅为3.24×106的情况下的峰值信噪比(PSNR)和结构相似性(SSIM)指标分别为28.53和0.9141,运行时间为0.3 s。

In order to improve the image deblurring restoration effect and processing speed,a motion blur removal algorithm based on deep convolutional neural network is proposed.According to the multi-scale image,a network model based on automatic encoder is constructed.In the aspect of expanding the receptive field,the dilated convolution module is used to extract the multi-scale feature information of the image and the residual module is used to widen the network depth and solve the problem of image detail loss during the training process,and the end-to-end motion blur removal task of the image is realized.The experimental results on the GOPRO data set and the real test set show that the peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)indicators of the algorithm proposed here reach 28.53 and 0.9141,and the running time is 0.3 s with the parameter amount of 3.24×106.