Comparing inpainting techniques for urban object restoration from orbital images

Abstract

Based on the comparison of three established inpainting techniques, namely Criminisi, Beltamio, and Galerne and Leclaire, our study aimed to identify the most effective method for road restoration after extraction and detection using a Mathematical Morphology operators combined with hybrid techniques of digital image processing in remote sensing images. While all techniques were evaluated based on both visual analysis and quantitative metrics, the Criminisi approach emerged as the better choice. Despite introducing some additional noise, this technique demonstrated superior performance in terms of Completeness and overall Quality, achieving approximately 95.23% and 94.56%, respectively. Its ability to accurately reconstruct linear geometries while effectively removing existing noise highlighted its suitability for road restoration tasks.