人脸图像超分辨率技术，又名人脸幻觉，可根据给定的低分辨率人脸图像中恢复出对应的高分辨率人脸图像。该技术无论是在学术界还是在工业界都具有非常广泛的应用前景。人脸，作为一种具有高度结构先验的对象，其结构先验可以为网络提供结构信息，从而辅助人脸图像超分辨率任务的执行，改善人脸图像超分辨率性能。因而许多基于结构先验的人脸图像超分辨率方法被提出。为了了解和掌握基于结构先验的人脸图像超分辨率技术的发展状况，本文对其进行了系统的总结与归类，主要从先先验、并行先验、中间先验和后先验，四个方面对基于结构先验的人脸图像超分辨率技术进行概述。最后分析基于结构先验的人脸图像超分辨率技术存在的问题与挑战。

Face super-resolution technology, also known as face hallucination, can recover the corresponding high-resolution face images from the given low-resolution face images. It has a very broad application prospect both in academia and industry. Human face is a highly structured abject, and its structural prior can provide the network structural information and then facilitates the execution of face super-resolution task and improve face super-resolution performance. In order to understand and master the development status of structural prior-based face super-resolution technology, this paper systematically summarizes and categorizes it. From pre-prior, parallel-prior, in-prior, post-prior, four aspects of the structural prior-based face image super-resolution technology are summarized. Finally, the paper analyzes the problems and challenges deep learning face super-resolution technology facing.