Дослідження методів обробки зображень. Методи реалістичного заміщення об'єктів

Abstract

This study explores advanced methods for realistic object substitution in digital images through machine learning techniques. Focusing on convolutional neural networks and generative models, it evaluates various algorithms including PatchMatch, GANs, DeepFill V2, and Stable Diffusion Inpainting, among others, for their effectiveness in seamless object integration. The analysis demonstrates the high efficiency of these methods in processing complex contextual relationships, ensuring that the integration of objects into original images appears natural and convincing.