Progressive face super-resolution with cascaded recurrent convolutional network

Abstract

Progressive upsampling is beneficial for deep learning based large factor (e.g., 8×) super-resolution (SR) to improve network performance and reduce the difficulties of network training. The feedback mechanism is helpful in strengthening the representation power of deep networks since it can efficiently enlarge the receptive field. In this paper, we propose a progressive cascaded recurrent convolutional network, dubbed PCRCN, for large factor face SR (FSR). Specifically, a novel multi-stage cascaded convolutional neural network is developed to progressively obtain high magnification face images, where the first stage of network achieves an initial 2× magnification image, and the following other stages, adopting the recurrent structure, sequentially generate the corresponding 4×,8× and possibly larger factor SR images through multiple independent iterative modules. The deep features and parsing priors of face are extracted in parallel in each stage of network, and integrated to improve the deep representation ability of network. The training of the whole network is supervised in an end-to-end way by the weighted sum of multiple losses. Compared with other state-of-the-art methods, the experimental results show that the proposed method can achieve superior results in terms of both subjective and objective evaluations.