The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-2/W4
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W4, 97–100, 2017
https://doi.org/10.5194/isprs-archives-XLII-2-W4-97-2017
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W4, 97–100, 2017
https://doi.org/10.5194/isprs-archives-XLII-2-W4-97-2017

  10 May 2017

10 May 2017

USING SUPERVISED DEEP LEARNING FOR HUMAN AGE ESTIMATION PROBLEM

K. A. Drobnyh and A. N. Polovinkin K. A. Drobnyh and A. N. Polovinkin
  • Lobachevsky State University of Nizhny Novgorod, Russia

Keywords: Machine Learning, Age Estimation, Supervised Deep Learning, Active Appearance Model, Bio-Inspired Feature, Support Vector Machine

Abstract. Automatic facial age estimation is a challenging task upcoming in recent years. In this paper, we propose using the supervised deep learning features to improve an accuracy of the existing age estimation algorithms. There are many approaches solving the problem, an active appearance model and the bio-inspired features are two of them which showed the best accuracy. For experiments we chose popular publicly available FG-NET database, which contains 1002 images with a broad variety of light, pose, and expression. LOPO (leave-one-person-out) method was used to estimate the accuracy. Experiments demonstrated that adding supervised deep learning features has improved accuracy for some basic models. For example, adding the features to an active appearance model gave the 4% gain (the error decreased from 4.59 to 4.41).