CvgTalk-Lulu-Periocular Recognition

最新推荐文章于 2024-08-04 12:41:36 发布

原创最新推荐文章于 2024-08-04 12:41:36 发布 · 607 阅读

1 ·

CC 4.0 BY-SA版权

CVG Talks 专栏收录该内容

2 篇文章

订阅专栏

本文探讨了一种结合3D形状与2D局部二值模式(LBP)的虹膜识别方法，旨在提高从远距离获取的虹膜图像的识别率。通过将3D面部模型与2D图像配准，利用ICP算法进行3D形状匹配，并计算探针与画廊中3D点云的距离匹配得分，最终实现了3D与2D纹理信息的融合。实验结果表明，该方法相较于纯2D纹理识别提高了超过20%的准确率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

                    
                    Iris region is small, esp. from distance
combine periocular region to increase detection region

Current status

mostly use ideal faces and iris images 

frontal face
high res. eye region image
issues  

only 2D

Proposed method

3D shape 
2D LBP 

to obtain 2D periocular region, first register 2D with 3D shape using facial landmark
3D shape matching, ICP
distance matching score between probed periocular 3D point cloud and that from gallery
fusion of 3D and 2D

Experiment

Bosphorus DB
3D face dataset
105 subject
4,666 face scans

Results

3D > 20% better than 2D texture
Intra-class variations