Position-Sensitive painterly image harmonization.

Abstract

Painterly image harmonization aims to insert photographic objects into paintings to create visually harmonious artistic images. Unlike realistic images where local variations are relatively uniform, paintings exhibit highly spatially dependent style variations due to brushstrokes, textures, and color blending, with potentially significant style differences across different regions. Existing methods have made efforts to align the style of photographic objects with the whole painting, but often neglect the local style consistency around the insertion position. To address this, we propose a Position-Sensitive Painterly Image Harmonization method. Specifically, we first propose a Position-aware Adaptive Instance Normalization (PAIN) module that integrates both global and local background styles in a position-sensitive way, ensuring the inserted object harmonizes with both the overall style and localized variations. Then, we propose a position-sensitive supervision method to detect local style inconsistencies and finally achieve position-sensitive painterly image harmonization. Extensive experiments on public benchmark datasets have validated the effectiveness of our method.