Super Resolution Using Convolutional Neural Network and Self-Attention Mechanism

The degradation of image features after upscaling, particularly the roughening of edges and curves, can significantly diminish both user experience and the effectiveness of the image's communicative intent. Enhancing image quality after increasing resolution remains a critical area of study. Traditional interpolation techniques, such as BICUBIC, although capable of generating larger images quickly, often result in suboptimal quality. To address this limitation, this research proposes the SRUNET model, which synergizes an attention mechanism with the UNET convolutional network architecture. The proposed model demonstrates the capability to process images of varying sizes effectively and presents a novel approach that yields competitive results when benchmarked against existing methodologies.