Human Age Estimation Using Ranking SVM

This paper proposes a human age estimation method using Ranking SVM method. Given a face image, most previous methods estimate the human age directly. However, the face images from the same age vary so much that it’s really a difficult problem to estimate the human age accurately. In this work, we adopt an alternative way to estimate the human age. First, the rank relationship of the ages is learned from various face images. Then, the human age is estimated based on the rank relationship and the age information of a reference set. There are two advantages of the proposed method. (i) The rank relationship rather than the absolute human age is learned so that the absolute age estimation problem can be simplified. (ii) The human age is determined based on the rank relationship and the known human age of the reference set so that the face image variations from the same age group can be considered. Experimental results on MORPH and Multi-PIE databases validate the superiority of the rank based human age estimation over some state-of-the-art methods.