A Comparative Study of Fractal Dimension Based Age Group Classification of Facial Images with Different Testing Strategies

The demand of estimation of age from facial images has tremendous applications in real world scenario like law enforcement, security control, and human computer interaction etc. However despite advances in automatic age estimation, the computer based age classification has become prevalent. The present paper evaluates the method of age group classification based on the Correlation Fractal Dimension (FD) of facial image using different validation techniques. To reduce variability, multiple rounds of cross validation are performed using different partitions to the data. The expected level of fit of the model classifying facial images into four categories based on FD value of a facial edge is estimated using multiple cross-validation techniques. The simulation is carried out and results are analyzed on different images from FG-NET database, Google database and from the scanned photographs as these are random in nature and help to indicate the efficiency and reliability of the proposed method. It is also a successful demonstration that Correlation Fractal Dimension of a facial edge is sufficient for a classification task with high percentage of classification accuracy.