An effective approach of facial age estimation with extreme learning machine

How to accurately estimate facial age is a difficult problem due to insufficiency of training data. In this paper, an effective approach is proposed to estimate facial age by means of extreme learning machine (ELM). In the proposed method, a set of features is randomly selected from the original features to consist of a feature subspace. Given an initial weight matrix, the training samples within the feature subspace are input to ELM to constitute a weaker estimator. Besides the feature subspace, the initial weight matrix is varied to construct multiple weaker estimators with a good diversity. In order to alleviate the negative affect caused by the sample imbalance of different ages, a weighting model is designed based on the training sample distribution. Experimental results on the standard database demonstrate the feasibility and effectiveness of the proposed method.