Facial expression recognition with FRR-CNN

Feature redundancy-reduced convolutional neural network (FRR-CNN) is proposed to address the problem of facial expression recognition. Different from traditional CNN, convolutional kernels of FRR-CNN is induced to be divergent by presenting a more discriminative mutual difference among feature maps of the same layer, which results in generating less redundant features and yields a more compact representation of an image. Furthermore, the transformation-invariant pooling strategy is used to extract representative features cross-transformations. Extensive experiments are conducted on two public facial expression databases and the obtained results demonstrate the efficiency of FRR-CNN comparing with the state-of-the-art expression recognition methods.