摘要： 图像在获取过程中受传感器、光学系统、大气扰动等多因素影响，导致图像分辨率下降。提出一种结合内外部梯度先验的图像超分辨率重建方法。基于最大后验估计理论框架，将深度学习的外部梯度先验和范数的内部梯度、亮度联合约束先验相结合，通过联合利用外部梯度先验和内部梯度先验的互补优势，借助高分辨率的梯度信息提高重建的超分辨率图像质量。实验结果表明，与ASAR、DRRN等方法相比，该方法能够有效增强图像的高频细节和边缘信息，减少重建图像中可能存在的噪声与伪影效果。

关键词: 超分辨率重建, 梯度先验, 联合约束, 深度学习, 最大后验概率

Abstract: Image acquisition is affected by sensors, optical systems, and atmospheric disturbances and so on, which often leads to the degradation of image resolution.This paper presents a novel super-resolution image reconstruction method under the joint constraints of external and internal gradient prior.Based on the framework of Maximum a Posterior Probability(MAP), the proposed method integrates the external gradient priors with the joint constraint prior of brightness and the internal gradient in norm.By combining the complementary advantages of external and internal gradient prior, the quality of reconstructed super-resolution images is improved by using the high-resolution gradient information.Results of simulation and real experiments show that compared with ASAR, DRRN and other methods, the proposed method can greatly enhance the high-frequency details and edge information, and reduce the possible noise and artifact effects in reconstructed images.

Key words: super-resolution reconstruction, gradient prior, joint constraints, deep learning, Maximum A Posterior (MAP) probability

中图分类号: 

• TP391