博文

AI复原”美男学霸”，一文综述背后神算法

已有 2118 次阅读 2019-8-19 09:43 |个人分类:好文推荐|系统分类:论文交流

·前言·

前不久，82岁院士王德民AI修复照爆红网络，《人民日报》赞其"美而不自知"。几十年前模糊的黑白老照片，通过AI处理，神奇地还原出一位科学界的"美男学霸"。

图片来自网络

所谓的AI复原，其实就是应用了超分辨率技术。什么是超分辨率？其涉及的模型与算法有哪些？研究前景如何？当超分辨率与深度学习技术相结合，又会带来哪些新思路？IJAC特约综述：基于深度学习的单个图像超分辨率，即日起限时免费下载，答案就在文中......

Deep Learning Based Single Image Super-resolution: A Survey

Viet Khanh Ha, Jin-Chang Ren, Xin-Ying Xu, Sophia Zhao, Gang Xie, Valentin Masero, Amir Hussain

全文下载：

1)SpringerLink:

https://link.springer.com/article/10.1007/s11633-019-1183-x

2)IJAC官网：

http://www.ijac.net/en/article/doi/10.1007/s11633-019-1183-x

当期(Vol.16, No.4)全部文章限时免费下载：

特约综述+最新研究，好文云集！

全文导读

单个图像超分辨率(Single image super-resolution)吸引了越来越多研究者的关注，并广泛应用于卫星成像(satellite imaging)、医疗成像(medical imaging)、计算机视觉(computer vision)、安全监控图像(security surveillance imaging)、遥感(remote sensing)、目标检测及识别(objection detection, and recognition)中。

所谓超分辨率(Single image super-resolution,SISR)，是指由低分辨率图像(low-resolution (LR) image)创建高分辨率图像(high-resolution (HR) images)的过程，这一技术可很好地用于解决各类现实问题，如处理因带宽(bandwidth)、像素大小(pixel size)、场景细节(scene details)及其他因素导致图像或视频不清晰的问题。

图片来自论文

总体而言，可以把能够解决SISR问题的各种技术分为三大类：基于插值的方法(interpolation-based)、基于重建的方法(reconstruction-based)和基于样本的方法(example-based methods)，三种方法各有优劣，文中进行了详细论述。

近年来，深度学习技术得到极大发展与繁荣，在很多领域催生出大量尖端科技，也给SISR带来了新思路。鉴于其在特征提取与映射(feature extraction and mapping)中表现出的良好性能，深度学习技术可很好预测低分辨率图像(low-resolution images)中丢失的高频细节(high-frequency details)。

图片来自论文

8月，IJAC出版了由英国思克莱德大学Jin-Chang Ren教授团队带来的特约综述：基于深度学习的单个图像超分辨率(Deep Learning Based Single Image Super-resolution: A Survey)。本文首先论述了研究背景，而后讨论了基于样本的各类SISR算法，并介绍了深度学习相关模型的最新研究进展，对比了不同基于卷积神经网络的SISR算法，最后展开深度讨论，提出未来研究可能的方向和需要解决的问题。

图片来自Springer

全文信息

Deep Learning Based Single Image Super-resolution: A Survey

Viet Khanh Ha, Jin-Chang Ren, Xin-Ying Xu, Sophia Zhao, Gang Xie, Valentin Masero, Amir Hussain

摘要：

Single image super-resolution has attracted increasing attention and has a wide range of applications in satellite imaging, medical imaging, computer vision, security surveillance imaging, remote sensing, objection detection, and recognition. Recently, deep learning techniques have emerged and blossomed, producing " the state-of-the-art” in many domains. Due to their capability in feature extraction and mapping, it is very helpful to predict high-frequency details lost in low-resolution images. In this paper, we give an overview of recent advances in deep learning-based models and methods that have been applied to single image super-resolution tasks. We also summarize, compare and discuss various models from the past and present for comprehensive understanding and finally provide open problems and possible directions for future research.

关键词：

Image super-resolution, convolutional neural network, high-resolution image, low-resolution image, deep learning.

全文下载：

1)SpringerLink:

https://link.springer.com/article/10.1007/s11633-019-1183-x

2)IJAC官网：

http://www.ijac.net/en/article/doi/10.1007/s11633-019-1183-x