一种基于深度学习的图像盲去运动模糊算法

An Image Blind Motion Deblurring Algorithm Based on Deep Learning

数字图像一直在信息的传递中扮演着重要的角色,但同时也存在着大量模糊图像的问题。无论是目标检测、自动驾驶还是人脸识别等计算机视觉任务都需要依托大量清晰的图像,因此将模糊图像重新变得清晰是一个急切且广泛的需求。针对因运动导致的图像模糊,本文工作设计一种新颖的基于深度学习的端到端的图像盲去运动模糊算法。算法通过结合监督式学习的思路,将对抗式生成网络模型进行结构改造,并通过针对性加入人脸先验信息设计的损失函数来引导网络学习图像对之间的差别,在恢复运动模糊图像方面取得显著的成果。

Digital images always play an important role in the transmission of information,but there are also plenty of blurry images.Computer vision tasks such as object detection,autonomous driving or face recognition require a large number of high-resolution images.Therefore,it is an urgent and widespread need to make blurry images deblurred.Aiming at the blurred images caused by motion,my work designs a novel end-to-end blind motion deblurring algorithm based on deep learning.By combining the idea of supervised learning,the algorithm modi⁃fies the structure of the generative adversarial network model,and help the model to learn the difference between image pairs under the guidance of the loss function with the prior information of the face.Significant results have been achieved in terms of restoring motion blurred images.