Table 3 Summary of age and age-group estimation studies
Publication | Feature representation | Face database/database size | Algorithm | Evaluation protocol | Performance/accuracy (MAE, CS) |
|---|---|---|---|---|---|
Kwon and Lobo 1999 [87] | Anthropometric model | Private/47 | Classification | 15 images for testing | CS 100% |
Kanno et al. 2001 [215] | Appearance model | Private/440 | Classification | N/A | CS 80% |
Lanitis et al. 2002 [66] | 2D shape, raw pixel values | Private/500 | Regression | 500 train, 65 test | MAE 4.3 |
Iga et al. 2003 [216] | GWT, 2D shape | Private/101 | Classification | N/A | CS 58.4% |
Lanitis et al. 2004 [23] | Active appearance model (AAM) | Private/400 | Classification and regression | 50% train, 50% test | MAE 3.82–5.58 |
Zhou et al. 2005 [217] | Anthropometric model | FG-NET/1002 | Regression | 5-fold cross-validation | MAE 5.81 |
Ueki et al. 2006 [134] | Anthropometric model | WIT-DB | Classification | 2-fold cross-validation | CS 50%(M), 43%(F) |
Takimoto et al. 2006 [218] | Anthropometric model | HOIP | Classification | Leave-one-out | CS 57.3%(M), 54.7%(F) |
Geng et al. 2006 [26] | AGES | FG-NET/1002 | Regression | Leave-one-person-out | MAE 6.77; CS 81% |
Takimoto et al. 2007 [219] | Anthropometric model | HOIP | Regression | Leave-one-out | MAE 3.0(M), 4.4(F) |
Gen et al. 2007 [13] | AGES | FG-NET/1002 MORPH | Regression | Train on FG-NET, Test on MORPH | MAE 8.83; CS 70% |
Yan et al. 2007 [220] | Active appearance model | FG-NET/1002 | Regression | Leave-one-person-out | MAE 5.33 |
Yan et al. 2007 [220] | Appearance model (raw image) | YGA/8000 | Regression | 4-fold cross-validation | MAE 6.95(M), 6.95(F); CS 79%(M), 78%(F) |
Yan et al. 2007 [140] | Active appearance model | FG-NET/1002 | Regression | Leave-one-person-out | MAE 5.78 |
Yan et al. 2007 [140] | Appearance model (raw image) | YGA/8000 | Regression | 1000 train, 3000 test | MAE 10.36 (M), 9.79 (F); CS 61%(M), 63%(F) |
Fu et al. 2007 [12] | Age manifold | YGA | Regression | 50% test, 50% train | MAE 8.0 (M), 7.8 (F); CS 70% (M), 70% (F) |
Yan et al. 2008 [83] | Appearance model (patches) | FG-NET/1002 | Regression | Leave-one-person-out | MAE 4.95 |
Yan et al. 2008 [83] | Appearance model (patches) | YGA/8000 | Regression | 1000 train, 3000 test | MAE 4.38 (M), 4.94 (F); CS 88% (M), 85% (F) |
Guo et al. 2008 [31] | Active appearance model | FG-NET/1002 | Regression | Leave-one-person-out | MAE 5.08 |
Guo et al. 2008 [145] | Active appearance model | FG-NET/1002 | Hybrid | Leave-one-person-out | MAE 4.97; CS 88% |
Guo et al. 2008 [145] | Age manifold | YGA/8000 | Hybrid | 4-fold cross-validation | MAE 5.12 (M), 5.11 (F); CS 83% (M), 82% (F) |
Yan et al. 2008 [84] | Appearance model (patches) | YGA/8000 | Regression | 1000 train, 3000 test | MAE 7.82 (M), 8.53 (F); CS 75% (M), 70% (F) |
Fu and Huang 2008 [12] | Age manifold | YGA/8000 | Regression | 50% test, 50% train | MAE 6.0 (M), 5.5 (F); CS 82% (M), 83% (F) |
Suo et al. 2008 [152] | Active appearance model, photometric | FG-NET/1002, private/8000 | Regression | N/A | MAE 6.45 (private), 6.97 (FG-NET) |
Zhuang et al. 2008 [221] | Appearance model (patches) | YGA/8000 | Regression | 50% test, 50% train | MAE 5.40 (M), 6.33 (F); CS 82% (M), 76% (F) |
Ni et al. 2009 [154] | Appearance model (patches) | MORPH, web images | Regression | Train on web images, test on MORPH | MAE 8.60 |
Yan et al. 2009 [192] | Active appearance model | FG-NET/1002 | Classification | Leave-one-person-out | MAE 5.21 |
Xiao et al. 2009 [222] | Active appearance model | FG-NET/1002 | Classification | Train 300, test 702 | MAE 5.04 |
Guo et al. 2009 [82] | Age manifold, BIF | FG-NET/1002 | Regression | Leave-one-person-out | MAE 4.77 |
Guo et al. 2009 [82] | Appearance model, BIF | YGA/8000 | Classification | 4-fold cross-validation | MAE 3.47 (M), 3.91 (F); |
| Â | Â | Â | Â | Â | CS 88% (M), 85% (F) |
Guo et al. 2009 [191] | Appearance model (BIF) + age manifold | YGA/8000 | Classification | 4-fold cross-validation | MAE 2.58 (M), 2.61 (F) |
Guo and Mu 2011 [123] | BIF | MORPH II/55,132 | Regression | 50% train, 50% test | MAE 4.2 |
Choi et al. 2011 [70] | Active appearance model, Gabor, LBP | FG-NET/1002, PAL/430, BERC/390 | Hybrid | Leave-one-person-out | MAE 4.7(FG-NET), |
| Â | Â | Â | Â | Â | 4.3(PAL), 4.7(BERC); |
| Â | Â | Â | Â | Â | CS 73%, 70%, 65%, |
| Â | Â | Â | Â | Â | respectively |
Luu et al. 2011 [197] | Contourlet appearance model | FG-NET/1002 PAL/443 | Hybrid | Leave-one-person-out | MAE 4.1 (FG-NET), |
| Â | Â | Â | Â | Â | 6.0(PAL); CS 74% |
| Â | Â | Â | Â | Â | (FG-NET) |
Chang et al. 2011 [195] | Active appearance model | FG-NET/1002 MORPH II/5492 | Classification | 20% test, 80% train | MAE 4.5(FG-NET), |
| Â | Â | Â | Â | Â | 6.1(MORPH); CS 74.4%, |
| Â | Â | Â | Â | Â | 56.3%, respectively |
Wu et al. 2012 [115] | Grassmann manifold of 2D shape | FG-NET/1002, passport/223 | Hybrid | Leave-one-person-out | MAE 8.84 (passport), |
| Â | Â | Â | Â | Â | 5.9 (FG-NET); CS 40% |
| Â | Â | Â | Â | Â | (passport), 62% (FG-NET) |
Thukral et al. 2012 [180] | Grassmann manifold of 2D shape | FG-NET/1002 | Hybrid | Leave-one-person-out | MAE 6.2 |
Chao et al. 2013 [194] | Active appearance model with distance metric adjustment | FG-NET/1002 | Regression | Leave-one-person-out | MAE 4.4 |
Lu and Tan 2013 [142] | Manifold of raw pixel intensities | MORPH II/20000 | Regression | 50% test, 50% train | MAE 5.2 (white), 4.2 |
| Â | Â | Â | Â | Â | (black); CS 67% |
| Â | Â | Â | Â | Â | (white), 59% (black) |
Hadid | Pietikainen 2013 [200] | Raw intensities, volume LBP | Internet videos/2000 | Classification | 10% test, 90% train; |
| Â | Â | Â | Â | Â | CS 83.1% |
Guo and Mu 2013 [124] | BIF | MORPH II/55132 | Regression | N/A | MAE 4.0 |
Geng et al. 2013 [201] | Active appearance model, label distribution | FG-NET/1002 MORPH II/55132 | Classification | Leave-one-person-out (FG-NET), 10-fold cross-validation (MORPH II) | MAE 4.8(FG-NET), 4.8(MORPH II) |
Han and Jain 2014 [223] | BIF | LFW/143, FG-NET/1002 | Classification | 98.8% train, 1.2% test | MAE 4.5(FG-NET); CS 66.7% (LFW), 68.1% (FG-NET) |
Han et al. 2015 [147] | Demographic informative features | FG-NET/1002 MORPH II/78, 207, PCSO/100,012, LFW/4,211 | Hybrid | N/A | MAE 3.8(FG-NET), 3.6(MORPH II), 4.1(PCSO), 7.8(LFW); CS 78% (FG-NET), 77.4% (MORPH II), 72.6% (PCSO), 42.5% (LFW) |
Nguyen et al. 2015 [141] | MLBP, Gabor | PAL/1160 | Regression | 50% test, 50% train | MAE 6.6 |
Li et al. 2015 [224] | 2D age manifold, BIF | FACES/1,026 FG-NET/1002 | Classification | N/A | MAE 1.3(FG-NET), 8.2(FACES) |
Onifade et al. 2015 [143] | Age-rank LBP | FG-NET/1002 | Regression | Leave-one-person-out | MAE 2.34 |
Yang et al. 2015 [225] | Raw intensities | Private/3615 | Classification | 2479 train, 1136 test | MAE 0.31 |
Huerta et al. 2015 [135] | HOG, LBP, SURF | FG-NET/1002 | Classification | Leave-one-person-out | MAE 3.31 |
Hu et al. 2016 [138] | Kullback-Leibler/raw intensities | LFW/150,000 MORPH II/55,132 FG-NET/1002 | Classification (CNN) | N/A | MAE 2.8(FG-NET), 2.78(MORPH) |
Akinyemi and Onifade 2016 [144] | Raw pixel intensities | FG-NET/1002 FAGE/238 | Regression | Leave-one-person-out | MAE 3.19 |