Multi-stage classification network for automatic age estimation from facial images

Existing age estimation algorithms based on facial images have been showing high dependency on the age range with the range 29-49 yielding the best estimation results. This paper introduces a new multi-stage binary age estimation (MSAE) system configured as a network of decision making neural network (NN) and support vector machine (SVM) units. The decision making process was based on the classification of image features derived by the Orthogonal Locality Preserved Projections (OLPP) and the Sobel Edge Detector (SED) algorithms. The proposed method was tested using the noncommercial version of the MORPH2 database. For male faces, the age estimation results for the MSAE method achieved above 90% of average accuracy for the 26-34 years of age range, and above 80% for 16-26 and 70% for 34-50 years of age. Similar trends were observed for female faces; however the accuracy was slightly lower due to smaller number of valid images.