Zero-sample mural superresolution reconstruction for enhanced perceptual quality

Abstract Aiming at the problem of texture loss and poor perceptual quality in low-resolution mural images, this paper proposes a zero-sample superresolution reconstruction method called EPZSSR to enhance quality, specific model is obtained by training image. The algorithm takes zero-shot as framework, randomly cuts original image into 128 * size, performs Gaussian blurring on image, uses Lanczos interpolation downsample smooth reduce artifacts, convolutional attention module skip connection optimize network structure. SmoothL1Loss used robustness model, PI value introduced evaluation index. experimental results show that compared with other algorithms, peak signal-to-noise ratio increased 0.98–3.23 dB average. effect better, reduced 0.56 average, perception running time 89.68 s It has certain for reconstruction.