SR-Net: A Super-Resolution Image Based on DWT and DCNN

Recently, a surge of several research interests in deep learning has been sparked for image super-resolution. Basically, a deep convolutional neural network is trained to identify the correlation between low and high-resolution image patches. In other side, profiting from the power of wavelet transform to extract and predict the “missing de-tails” of the low-resolution images, we propose a new deep learning strategy to predict missing details of wavelet sub-bands in order to generate the high-resolution image which we called a super-resolution image based on discrete wavelet transform and deep convolutional neural network (SR-DWT-DCNN). By training various images such as Set5, Set14 and Urban100 datasets, good results are obtained proving the effectiveness and efficiency of our proposed method. The reconstructed image achieves high resolution value in less run time than existing methods based on based on the evaluation with PSNR and SSIM metrics.