基于深度编码–解码器的图像模糊核估计

Blur kernel estimation method based on deep encoder-decoder network

中文关键词  图像去模糊  编码-解码器  显著边缘  卷积核估计

英文关键词  Image Deblurring  Encoder-decoder network  Significant-edges  estimation of blur kernel

中文摘要

      在图像去模糊问题中，显著边缘结构对图像的模糊核估计具有重要的作用。本文提出一种基于深度编码-解码器的图像模糊核估计算法。首先，通过构建训练数据集对深度编码-解码器进行训练，进而自适应地获得模糊图像的显著边缘结构；接着，结合显著边缘结构和模糊图像，利用$L_2$范数正则化对模糊核进行估计；最后，利用超-拉普拉斯先验和所估计的模糊核对清晰图像进行估计。与传统的方法相比，所提出的方法不需要多尺度迭代框架。实验结果表明，所提出的算法在获得较好的显著边缘结构以及清晰图像的同时，能够减少算法计算的时间

英文摘要

      The significant edges-structure of blurry image plays an important role in estimation of blur kernel with the image deblurring problem. This paper proposes a method for estimating blur kernel based on deep encoder-decoder network. Firstly, after training by the constructing train dataset，the deep encoder-decoder network obtains the significant-edges of blurry image adaptively. And then, the blur kernel can be estimated by the $L_2$ norm regularization where combining the significant edge-structures and the blurry image. Finally, the latent image can be estimated by using the hyper-Laplacian priors after estimating blur kernel. Comparing with the traditional methods, the proposed method does not use the multi-scale iterative framework to estimate kernel. Experimental results show that the proposed method can reduce the computation time of the algorithm while obtaining better significant edge-structures and latent image.