Multiscale Template Matching to Denoise Epigraphical Estampages

An epigraph is an engraved inscription from the past. Deciphering, recognizing and understanding the text in such epigraphical documents is mundane work for archaeologists. One of the major tasks in this is preprocessing the document and denoising it into segmentable form in an automated and fast manner. In this paper, we propose a pipelined solution which includes denoising techniques such as nested median filtering followed by a method of multiscale template matching to denoise noisy epigraphical documents. All epigraphs are usually in grayscale and they are converted to binary black and white images using otsu thresholding. Our template matching algorithm has been enhanced to create a mask of all the matched noise. A deep learning technique called Image Inpainting is then used in order to remove the noise and replace it with the respective background. A comparative study is performed with the output of the proposed method with use of novel denoising methods such as Gaussian blur and wiener filters. Human perception can determine that the proposed algorithm provides more satisfactory results. Standard quality measures such as the Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) are used to show that the integrity of the image with respect to the text remains intact.