Efficient Image Super-Resolution Integration

The super-resolution (SR) problem is challenging due to the diversity of image types with little shared properties as well as the speed required by online applications, e.g., target identification. In this paper, we explore the merits and demerits of recent deep learning-based and conventional patch-based SR methods and show that they can be integrated in a complementary manner, while balancing the reconstruction quality and time cost. Motivated by this, we further propose an integration framework to take the results from FSRCNN and A+ methods as inputs and directly learn a pixel-wise mapping between the inputs and the reconstructed results using the Gaussian conditional random fields. The learned pixel-wise integration mapping is flexible to accommodate different upscaling factors. Experimental results show that the proposed framework can achieve superior SR performance compared with the state of the arts while being efficient.