Evaluation of Traditional and Modern Inpainting Algorithms for Image Restoration
Aditya Kurniawan, Kirsten Winther Thjahyadi, Vinny Surya Trinita · 2024
Image restoration remains a critical area of study in digital forensics and computer vision, requiring dependable tools to reconstruct missing or damaged parts of images. Four prominent inpainting algorithms, such as OpenCV's Navier-Stokes, OpenCV's Telea, Laplacian, and bi-Laplacian will be discussed and evaluated for their performances. The experiment is designed with controlled conditions and generating diverse data scenarios, such as removing and writing over images in various degrees of damages. A few changes made to the code script are also reviewed. These methods will be assessed based on the calculation results of several key metrics, which includes Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Histogram Image Similarity. This paper suggests that traditional inpainting algorithms remain fundamental due to their consistency and efficiency.