A Study of Convolutional Sparse Feature Learning for Human Age Estimate

Human age estimation plays an important role inhuman facial image analysis. Aging feature representation isone of the widely studied problems in this topic. Convolutionalmap (bio-inspired features, or BIF) has been proven to be themost successful framework, but its manual crafted filters cannot easily capture the complicated facial aging pattern. In thispaper, we adopt this convolutional map framework but proposea novel feature learning approach based on convolutional sparsecoding (CSC) that can automatically learn to characterizeaging signatures. Compared to other popular feature learningapproaches like deep convolutional neural network (CNN), weverify that our learning approach can extract localized subtleaging features like CNN, and also significantly reduce themodel size. Moreover, we employ the standard deviation (STD)pooling to summarize the aging feature. Finally, the extractedfeatures are fed into a discriminative manifold learning modelto obtain more discriminative low-dimensional representationsand further improve the computational efficiency. We evaluateour approach over the standard benchmark datasets. Theexperimental results demonstrate that our approach impressivelyoutperforms the state-of-the-art results. The proposedage estimation scheme also performs well in the cross-databaseage estimation task.