Human Age Estimation Based on Bio-Content Features on Facial Images via Gender Based Classification

Abstract

Age estimation is achieved better using facial images. Normally, the age estimation is categorized in two ways such as age estimation and age group classification. The age estimation is performed to estimate the exact age of a given test image. Whereas the age group classification is done to classify the age group of a person such as adults, junior adults and senior adults. The age group classification is supported by the features and patterns which is clearly defined in craniofacial growth. The wrinkles and ratios play an important role in age group classification. These similar features can be also applied for age estimation of an individual. Our proposed methodology estimates the age of a person in three steps. 1. Face detection, 2. Feature extraction, 3. Gender based classification, 4. Age estimation.

Face detection is achieved using Viola-Jones algorithm. After pre-processing and contract enhancement, the proposed facial feature called Bio-Content Feature (BCF) is extracted. The proposed feature is extracted using angle-based and dimension-based measurements and dimension reduction methods. The proposed extraction methodology handles the problems like image misalignment, high dimensionality problems, shape variations and geometrical transformations.

The extracted feature is subjected to multi-linear regression to find the relationship between the test and the trained image. The third step performs gender based classification of the images based on the result of the regression. The fourth step delivers the estimated age of the test image. This proposed work shows its accuracy up to 92% via gender-based classification on age estimation. The advantage of the proposed methodology for gender-based classification is proved by vast experiments on the public available FG-NET database. The approach could be widely used in real world applications, crime investigation, and human-computer interaction.