PROJECT TITLE :

DCSR Dilated Convolutions for Single Image Super-Resolution

ABSTRACT:

Expanding the receptive field without the loss of resolution or exploration of parameters is possible with dilated convolutions. Multiscale single image super-resolution (SR) using dilated convolutions is proposed in this study. In order to increase the receptive field size without increasing computational complexity, we use dilated convolutions. Each layer is called a mixed convolution, and the feature extracted by dilated convolutions and conventional convolutions is concatenated in the mixed convolution layer. Mixed convolutions' receptive field and intensity are theoretically analysed to determine their role in SR. A decent generalisation ability may be achieved by removing blind spots and capturing the correlation between LR and HR picture pairings using mixed convolutions. By training 5-layer and 10-layer networks, we demonstrate the properties of mixed convolutions. Additionally, we train a 20-layer deep network to compare the proposed method's performance to that of the current state of the art. The network also learns maps of different scales from an LR image to its HR one simultaneously, resulting in a more accurate model. Results show that our method outperforms current methods in terms of both PSNR and SSIM, particularly when applied to large-scale factors.