Deep Convolutional Networks-Based Image Super-Resolution

Convolutional neural networks (CNN) have been successfully applied in many fields of image processing, such as deblurring, denoising and image restoration. Estimating a high quality high-resolution image from one or a set of low-resolution images is a non-linear mapping, which can be formulated as a regression problem. According to the image formation process, a Deep Convolutional Network-based image Super-Resolution model DCNSR is proposed and is trained using end-to-end. Several key components of DCNSR, which would affect the training time and the effectiveness of reconstruction super-resolution image, are firstly demonstrated. Then, the deblurring performance is evaluated. Finally, comparisons with the results in state-of-the-arts are presented. Experimental results demonstrate that the proposed model achieves a notable improvement in terms of both quantitative and qualitative measurements.