APSIPA Transactions on Signal and Information Processing > Vol 13 > Issue 1

Deep Learning for Face Super-Resolution: A Techniques Review

Bolin Zhu, Wuhan Institute of Technology, China, Kanghui Zhao, Wuhan Institute of Technology, China, Tao Lu, Wuhan Institute of Technology, China, lutxyl@gmail.com , Junjun Jiang, Harbin Institute of Technology, China, Zhongyuan Wang, Wuhan University, China, Kui Jiang, Harbin Institute of Technology, China, Zixiang Xiong, Texas A&M University, USA
 
Suggested Citation
Bolin Zhu, Kanghui Zhao, Tao Lu, Junjun Jiang, Zhongyuan Wang, Kui Jiang and Zixiang Xiong (2024), "Deep Learning for Face Super-Resolution: A Techniques Review", APSIPA Transactions on Signal and Information Processing: Vol. 13: No. 1, e36. http://dx.doi.org/10.1561/116.20240045

Publication Date: 12 Dec 2024
© 2024 B. Zhu, K. Zhao, T. Lu, J. Jiang, Z. Wang, K. Jiang and Z. Xiong
 
Subjects
Image restoration and enhancement,  Deep learning,  Face detection and recognition,  Feature detection and selection,  Image and video processing
 
Keywords
Face super-resolutiondeep learningsurveyface characteristics
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 106 times

In this article:
Introduction 
Problem Settings and Terminology 
FSR Technologies and Methods 
Current Issues and Future Work 
References 

Abstract

Face Super-Resolution (FSR) represents a significant branch of image super-resolution, aiming to reconstruct low-resolution face images into high-resolution counterparts. Recently, driven by rapid advancements in deep learning technology, FSR methods using deep learning have achieved notable subjective and objective reconstruction quality, attracting extensive industrial attention. However, detailed classifications of FSR methods remain limited. Therefore, this survey systematically and comprehensively reviews deep learning-based FSR methods. Initially, we introduce the background and technical framework of FSR. Subsequently, we detail the FSR problem definition, alongside commonly used datasets, evaluation metrics, and loss functions. We conduct comprehensive researches in deep learning FSR methods and classify them according to their solution strategies. Within each category, we begin with a general method description, and subsequently introduce representative approaches and discuss their respective pros and cons. Finally, we address current challenges in FSR methods and propose future research directions.

DOI:10.1561/116.20240045