Enhanced ZSSR for Super-resolution Reconstruction of the Historical Tibetan Document Images
Yuqi Lu, Tiejun Wang, Penghai Zhao, Guowei Zhang · 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) · 2022
Due to the poor preservation and imaging conditions, the image quality of historical Tibetan document images is relatively unsatisfactory. In this paper, we adopt super-resolution technology to reconstruct high quality images of historical Tibetan document. To address the problem of low quantity and poor quality of historical Tibetan document images, we propose the EZSSR network based on the Zero-Shot Super-resolution Network (ZSSR), which borrows the idea of feature pyramid in Deep Laplacian Pyramid Networks (LapSRN) to extract different levels of features while alleviating the ringing artifacts. EZSSR neither requires paired training datasets nor preprocessing stage. The computational complexity of EZSSR is low, and thus, EZSSR can also reconstruct image within the acceptable time frame. Experimental results show that EZSSR reconstructs images with better visual effects and higher PSNR and SSIM values.