摘要
光照问题是人脸识别系统中的一个瓶颈问题。将常用于光照补偿的图像处理方法和基于模型的光照锥方法进行了对比实验,实验结果表明光照锥方法具有更好的补偿效果。同时对在构建光照锥时,如何选择训练图像问题进行了实验研究,结果表明应选择光照方向变化适中的图像,构建的光照锥光照补偿效果可以达到最佳,人脸的平均识别率可以提高1～6个百分点。
Illumination is a bottle-neck problem in face recognition. The contrastive experiments between image processing methods usually applied for illumination compensation and the illumination cone method based on model are performed. The results indicate that the illumination cone method is better. At the same time, the experimental research on how to choose training images to establish the cone is carried out. It is demonstrated that if the images in which the illumination direction varies moderately are chosen, the efficiency of illumination compensation can be up to the best, and the average recognition ratio can be improved 1-6 percent.
出处
《工程图学学报》
CSCD
北大核心
2009年第3期113-120,共8页
Journal of Engineering Graphics
关键词
计算机应用
人工智能
人脸识别
光照锥
光照
computer application
artificial intelligence
face recognition
illumination cone
illumination
