基于多方向韦伯梯度直方图的人脸识别 
下载： 1038次
Face Recognition Based on Multi-Directional Weber Gradient Histograms
湘潭大学物理与光电工程学院, 湖南 湘潭 411105
图 & 表
图 1. 3×3邻域图
Fig. 1. 3×3 neighborhood map
下载图片 查看原文
图 2. 计算过程。 (a)原始编码;(b)噪声编码1;(c)噪声编码2
Fig. 2. Counting process. (a) Original coding; (b) noise coding 1; (c) noise coding 2
下载图片 查看原文
图 3. MWGH算法流程图
Fig. 3. Flow chart of MWGH algorithm
下载图片 查看原文
图 4. YALE人脸库
Fig. 4. YALE face database
下载图片 查看原文
图 5. AR人脸库。(a)训练样本;(b)光照集;(c)表情集;(d)遮挡集A;(e)遮挡集B
Fig. 5. AR face database. (a) Training sample; (b) illumination subset; (c) facial expression subset; (d) partial occlusion subset A; (e) partial occlusion subset B
下载图片 查看原文
图 6. ORL人脸库
Fig. 6. ORL face database
下载图片 查看原文
图 7. 不同人脸库在不同分块方式下的识别率。(a) YALE人脸库;(b) AR人脸遮挡B库;(c) ORL人脸库
Fig. 7. Recognition rates for different face databases in different block modes. (a) YALE face database; (b) AR face database with partial occlusion subset B; (c) ORL face database
下载图片 查看原文
图 8. 不同人脸库在不同分块方式下的识别率
Fig. 8. Recognition rates for different face databases in different block modes
下载图片 查看原文
表 1YALE人脸库上的识别率
Table1. Recognition rates for YALE face database%
| Algorithm | Sample number |
|---|
| 1 | 2 | 3 | 4 | 5 |
|---|
| HOG | 94.00 | 94.67 | 84.67 | 93.33 | 95.33 | | WLD | 93.33 | 89.33 | 88.00 | 96.67 | 95.33 | | WLBP | 94.67 | 96.00 | 93.33 | 98.00 | 98.33 | | HWOG | 95.00 | 97.00 | 95.67 | 99.00 | 98.67 | | IGLBP | 96.67 | 98.00 | 93.33 | 99.33 | 98.67 | | DWLD | 94.33 | 95.33 | 92.00 | 98.00 | 96.00 | | MWGH | 98.00 | 99.33 | 97.33 | 99.33 | 99.33 |
|
查看原文
表 2AR人脸库上的识别率
Table2. Recognition rates for AR face database%
| Algorithm | Illuminationsubset | Expressionsubset | Partial occlusionsubset A | Partial occlusionsubset B |
|---|
| HOG | 91.33 | 90.33 | 69.00 | 49.00 | | WLD | 91.67 | 92.00 | 90.67 | 78.67 | | WLBP | 94.00 | 95.33 | 91.67 | 80.00 | | HWOG | 95.33 | 95.00 | 94.00 | 85.67 | | IGLBP | 99.67 | 99.00 | 98.67 | 94.33 | | DWLD | 93.00 | 94.67 | 92.00 | 83.67 | | MWGH | 99.33 | 100.00 | 99.33 | 94.67 |
|
查看原文
表 3ORL人脸库上的识别率
Table3. Recognition rates for ORL face database%
| Algorithm | Sample number |
|---|
| 2 | 3 | 4 | 5 |
|---|
| HOG | 87.94 | 92.36 | 94.17 | 96.00 | | WLD | 87.14 | 92.14 | 95.23 | 96.80 | | WLBP | 90.20 | 93.82 | 96.26 | 97.12 | | HWOG | 91.38 | 95.74 | 98.15 | 98.61 | | IGLBP | 90.81 | 94.46 | 96.50 | 97.95 | | DWLD | 89.35 | 94.26 | 96.66 | 97.20 | | MWGH | 93.90 | 97.21 | 98.80 | 99.26 |
|
查看原文
表 4在AR光照库上的加噪实验结果
Table4. Results from noise-added experiment for AR light database%
| Method | Normalized variance of Gaussian white noise | ψ |
|---|
| 0 | 0.0001 | 0.0002 | 0.0003 | 0.0004 |
|---|
| HOG | 91.33 | 64.73 | 49.20 | 40.20 | 33.73 | 63.07 | | WLD | 91.67 | 81.67 | 54.27 | 37.13 | 25.47 | 72.22 | | WLBP | 94.00 | 85.25 | 60.33 | 47.80 | 39.38 | 58.11 | | HWOG | 95.33 | 88.75 | 65.43 | 51.28 | 46.55 | 51.17 | | IGLBP | 99.67 | 93.33 | 90.67 | 85.33 | 80.00 | 19.74 | | DWLD | 93.33 | 91.87 | 86.67 | 75.93 | 64.20 | 31.21 | | MWGH | 99.33 | 97.07 | 92.67 | 80.40 | 69.87 | 28.94 |
|
查看原文
表 5不同算法在YALE人脸库的特征维数与耗时
Table5. Feature dimension and time-consuming of different algorithms on YALE face database
| Method | Featuredimension | T1 /ms | T2 /ms |
|---|
| HOG | 810 | 2.8 | 3.3 | | WLD | 12000 | 14.3 | 4.7 | | WLBP | 11800 | 10.5 | 4.7 | | HWOG | 1322 | 8.3 | 3.4 | | IGLBP | 16384 | 33.4 | 4.9 | | DWLD | 12000 | 14.5 | 4.7 | | MWGH | 13080 | 17.2 | 4.8 |
|
查看原文
杨恢先, 徐唱, 曾金芳, 陶霞. 基于多方向韦伯梯度直方图的人脸识别[J]. 激光与光电子学进展, 2018, 55(11): 111008. Huixian Yang, Chang Xu, Jinfang Zeng, Xia Tao. Face Recognition Based on Multi-Directional Weber Gradient Histograms[J]. Laser & Optoelectronics Progress, 2018, 55(11): 111008.
PDF全文
