Abstarct: Abstract In recent years, image super-resolution has emerged as a key research area in the fields of image processing and computer vision. The main objective of this technique is to reconstruct high-resolution images from low-resolution input images. Although deep neural networks have led to significant advancements in this domain, several challenges such as large number of model parameters and high computational complexity still remain. To address these issues, this thesis proposes three innovative methods named RPCAN, RC-SPAN, and RPC-SPAN. These approaches are implemented baxsed on the baxseline RCAN model. In RPCAN, a novel parallel channel attention unit incorporating contrast and global average pooling across parallel branches is introduced, which assigns higher weights to channels containing more detailed information. This approach improves both quantitative evaluation metrics and the visual quality of the reconstructed images. For instance, the experimental result for applying on Urban100 dataset in scale 4 shows 0.45% increasing PSNR and 0.38% increasing the SSIM respect to the RCAN model. In RC-SPAN, by designing parallel convolutional laxyers, the number of model parameters is reduced (by nearly 47%) without notably compromising the image reconstruction quality. Finally, the RPC-SPAN method combines the advantages of the previous two approaches, achieving improved image quality while maintaining a low number of model parameters. The proposed methods have been evaluated using widely used benchmark datasets including Set5, Set14, BSD100, Urban100, Manga109, and Historical, and tested across various upscaling factors. The results demonstrate substantial improvements in both quantitative metrics and visual quality, particularly considering the reduction in model parameters and computational complexity. Moreover, visual evaluations indicate more accurate recovery of edges and textures in the output images. Additionally, the notable reduction in model size and computational complexity in RC-SPAN and RPC-SPAN makes it feasible to deploy these intelligent models on resource-constrained devices such as mobile phones.