Técnicas de processamento de imagens para auxiliar na geração de mascaras para Inpainting

Abstract

Image restoration is a classic low-level problem in computer vision. Image restoration methods aim to undo or remedy defects in images, restoring low-quality images. One way to address these effects through software is by using an inpainting algorithm. Many of these algorithms require not only an input image, but also a mask that marks the regions where the algorithm should perform the inpainting. The mask can be created manually or obtained through the use of an algorithm such as a neural network. In this work, we demonstrate how digital image processing techniques can be used as a less costly alternative for creating masks for inpainting. We use techniques for image segmentation, such as thresholding and clustering, median filtering for noise reduction, and morphological operations, both for detail enhancement and removal of others. The masks were used with the RePaint inpainting method to test their quality and compared with masks generated by a scratch detection network. Ultimately, the results obtained suggest that the use of image processing techniques can generate masks that lead to good inpainting results. Keywords: Image processing. Mask. Inpainting.