Exploiting Monocular Depth Estimation for Style Harmonization in Landscape Painting
Sungho Kang, Hyunkyu Park, YeongHyeon Park, Yeonho Lee, Hanbyul Lee, Seho Bae, Juneho Yi · 2023
Style harmonization, also known as painterly image harmonization, is a technique to seamlessly blend objects from realistic photos into the background of paintings of different styles. While deep learning-based methods have produced satisfactory results, there are important factors to consider for successful style harmonization depending on the type of image being targeted. In particular, if a certain region of the target image takes up a large portion of the entire image, the object to be blended is largely affected by the region, failing to achieve plausible results of object conversion synthesis. In this study, focusing on landscape painting in which a significant portion of the entire image is occupied by sky regions, we proposed a novel method dubbed Dominant Region Discarding (DRD) that effectively removed the sky region with the monocular depth estimation and Segment Anything Model (SAM). The experimental results showed that the proposed method outperformed previous style harmonization methods.