Image restoration algorithms (inpainting): classical and neural network approaches

Abstract

The paper considers inpainting algorithms used to restore damaged or deleted image fragments in modern graphic editors. The evolution of approaches to filling missing areas is analysed, from classical methods based on differential equations (PDE) to patch-based algorithms and modern neural network models. The features of each approach, their advantages and limitations when processing textures, structural elements and complex scenes are identified. Particular attention is paid to the application of inpainting in intelligent image editing tools, in particular, automatic object removal and background reconstruction. It is shown that the implementation of deep learning methods has significantly improved the quality and realism of restored images, making this technology one of the key areas of development for modern graphic editors.