Image text deblurring by convolutional neural networks
Ali Shakir Alahmed, Serkan Özbay · AIP conference proceedings · 2023
In this research paper, a modern autoencoder architecture based on convolution neural networks is suggested for document deblurring. The main goal of this project is to achieve image deblurring without having any prior knowledge about the blur kernel. The approach creates a sharp image from a blurry one, demonstrating the utility of convolutional neural networks in document deblurring. Only a blurred image is used as input to find a sharp image. As a result, the blur kernel knowledge isn’t requested. In the evaluation step, the predicted image is assessed by using peak signal to noise ratio (PSNR) and structural similarity index (SSIM). The results indicate a major increase in visual quality over the studies revelated using a dataset of text images.