Extracting age features and age detection from facial skin images (part 2)

• Main Author: Tay, Sean Guoliang.
• Other Authors: Sung, Eric
• Format: Final Year Project
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics
• Online Access: http://hdl.handle.net/10356/17950
• Institution: Nanyang Technological University

Humans are generally able to estimate a person to a precise age group by looking at facial features. For computers to encompass such visual capabilities would indicate a major viable advantage and would probably be constructive in a lot of applications. Till now, computer vision is unable to accomplish what is deemed by exact age detection. In this project, the students work on the assumption that elderly skin patches contain larger components of higher frequencies due to their wrinkle geography on the faces. An algorithm will be written to extract skin patches; forehead, sides of eye and cheeks, from the digital images which contain aging features. Another algorithm is written to allocate these patches through Principal Components Analysis (PCA) to obtain the Eigenvalues and Eigenvectors. Euclidean Distance will then be obtained by performing simple arithmetic calculations when new vectors are derived. The classification of the age group is determined by the K-Nearest Neighbour Rule comparing the Euclidean Distance obtained from the Principal Components Analysis. For this project, four algorithms were written in total to test the accuracy of PCA in the classification of age groups. These algorithms were executed to an appropriate number of times to test for their consistency. Results generated from the different algorithms were tabulated and they showed close resemblance with regards to the accuracy and consistency. It was clear that age classification through PCA at the end of the project is able to achieve an average accuracy of 70 percent.