Research Of Age Estimation Method Based On Facial Images

Automatic age estimation based on facial images is a hot but challenging topic inHuman-Computer Interaction. The research of age estimation has both theoretical andpractical significance, and it could be applied to various application fields such asidentifying teenager, image and video retrieval, advertisement investigation, etc．In this paper, a new age estimation method based on facial images is proposed,investigating both age-relevant feature extraction and age estimation modeling. Theage-relevant feature extraction involves two types of features, facial geometric proportionfeatures and facial local texture features. Firstly, we establish a face anthropometrytemplate based on craniofacial growth pattern theory, and calculate the correlationcoefficients between each geometric proportion and the age value. Based on thecorrelation coefficients, we select a set of facial geometric proportion features which havethe strongest correlation with age variation. Secondly, we use fractional differentialapproach to extract facial local texture features of facial areas bounded by some featuredpoints on face. Finally, combine these two kinds of features to form personal age featurevectors. At the age estimation modeling stage, we use machine learning algorithms toobtain age-feature knowledge matrix, and design a voting strategy based on k-NNclassification algorithm to select the most nearest k training samples as candidates for aninput testing sample according to each feature. Then the age estimation output for theinput testing sample is decided by some statistics of these candidates.Out method could estimate the age value as well as the age intervals for an inputtesting facial image. Experimental results show that the estimation error is small and theclassification accuracy is close to human judgment.