仿真假体视觉下的人脸识别研究进展

The Research Progress of Facial Recognition Under Simulated Prosthetic Vision

目的:综述基于仿真假体视觉的人脸识别研究的主要进展。方法:回顾了近年来各研究小组基于仿真假体视觉的人脸识别研究进展,讨论假体视觉下人脸识别的可能性,分析了假体视觉下人脸识别的影响因素。结果:在仿真假体视觉下,分辨率和缺失率是影响人脸识别的主要因素,分辨率增加或是缺失率减小,对人脸的识别率有显著提高;其它光幻视参数中,一定程度内的灰度提高对假体下的人脸识别率有显著提高,而点大小、点间隙、对比度对其影响稍小,正方形的光幻视阵列排布的识别率高于六边形和极坐标排布。另外,图像处理策略如基于感兴趣区域的放大和对比度增强等,能帮助被试进行人脸识别。结论:由心理物理学方法得出,在假体视觉下,仍能完成基本的人脸识别任务;改变光幻视的参数或图像处理策略,会影响识别率。这些结果将帮助研究者们优化视觉假体中的信息处理和图像处理策略,并为术后康复训练提供了实验理论依据。

Objective： This paper presents a review on the current research progress of facial recognition under simulated prosthetic vision.Methods： By reviewing the latest progress of many research groups on facial recognition under simulated prosthetic vision,the possibility of accomplish this task is discussed and the factors that affect recognition are stated as well as analyzed.Results： Resolution and drop-out rate are the main factors that affect facial recognition under simulated prosthetic vision.The increase of resolution or the decrease of drop-out rate improves the recognition accuracy significantly.Among other phosphene parameters,the increase of grey level within certain range improves the recognition accuracy significantly.For the rest of them such as dot size,gap size and contrast,the effects are less significant.The rectangular grid is better than hexagonal gird and log-polar grid for facial recognition.Certain image processing strategies such as region-of-interest digital zoom and contrast enhancement can help recognition.Conclusions： Through psychophysical evaluation,conclusions can be drawn that facial recognition is possible under prosthetic vision and the performance is affected by phosphene parameters as well as image processing strategies.This will help the designer of visual prosthesis to optimize the information processing and image processing strategies,and give theoretical bases towards postoperation rehabilitation regime.