摘要
可见光图像易受光照变化影响,而热红外图像对成像的光照条件具有鲁棒性,因此,热红外图像可以弥补可见光图像光照敏感性这一不足。然而,红外热像仪价格昂贵,采集热红外图像的成本远高于可见光图像。针对此问题,提出了一种基于生成对抗网络的热红外人脸图像生成方法,采用条件生成对抗网络结合L1损失从可见光图像中生成红外热像。在USTC-NIVE数据库上的实验结果验证了所提出的红外热像生成方法的有效性。同时,将生成的红外热像作为扩充样本,有助于提高红外表情识别的精度。
Visible light images are sensitive to illumination change,while thermal infrared images are robust to light conditions. Therefore,thermal infrared images can make up for visible light images＇ weakness. However,infrared thermal camera is very expensive,which makes thermal infrared images not as available as visible light images. Thus,in this paper,we propose a method for generating thermal infrared images from visible light images by combining conditional generative adversarial networks with L1 loss. The results on USTC-NVIE database prove the feasibility and effectivity of this method. At the same time,using the generated images as expansion of samples,the performance on recognition task is improved.
作者
王雅欣
史潇潇
Wang Yaxin1, Shi Xiaoxiao2(1. School of Software, University of Science and Technology of China, Hefei 230015, China; 2. School of Computer Science and Technology, University of Science and Technology of China, Hefei 230027, Chin)
出处
《信息技术与网络安全》
2018年第8期40-44,共5页
Information Technology and Network Security
关键词
生成对抗网络
图像生成
表情识别
generative adversarial networks
image generation
expression recognition
