改进超分辨率卷积神经网络和字典学习的图像超分辨率重构算法

Super-resolution Reconstruction Algorithm Based on Improved Super-resolution Convolutional Neural Network and Dictionary Learning

为了提高图像超分辨率重构效果,在保留结构化信息的同时弥补高、低分辨率图像之间的高频信息损失,本文结合深度学习方法,提出了一种基于改进超分辨率卷积神经网络和字典学习的图像超分辨率重构方法.首先使用卷积神经网络所学习到的结构化的图像信息训练一个端到端的图像超分辨率重构模型,再采用字典学习模型对图像残差部分的高频信息进行补偿,从而获得具有更好视觉效果和峰值信噪比的高分辨率图像.实验结果表明,本文算法在主观评价上比其他代表性算法更加清晰并且较好的恢复了高频信息,提高了峰值信噪比值.

In order to improve the image super-resolution reconstruction effect,while preserving the structured information while making up the high-frequency information loss between high and low resolution images,this paper proposes an image super-resolution reconstruction method based on improved SRCNN and dictionary learning.First,an end-to-end image super-resolution reconstruction model is trained using the structured image information learned by the convolution neural network.And then use the dictionary learning model to compensate for the high frequency information of the image residual part,obtain a high resolution image with better visual effect and peak signal to noise ratio （PSNR）.The experimental results show that the proposed algorithm is more clear and better in the subjective evaluation than other representative algorithms,and the PSNR value is higher.