Monolithic image decomposition

Abstract Image decomposition is one of the essential image processing techniques in computer vision and computational photography because it can be applied to various areas, such as image smoothing, detail enhancement, image abstraction, and high-dynamic-range compression. The main goal of image decomposition is to successfully separate structure from a given image by preserving edge-like structure components and removing fine-scale details without prior information. This paper proposes an effective image decomposition technique called monolithic image decomposition that considers both local and global features using RGB color channels simultaneously by exploiting low-rank approximation and total variation-based minimization. Our approach is different to previous approaches in that previous approaches use either local or global features, and perform the image decomposition process channel by channel and combine the decomposition result of each channel. Using monolithic parameter update, we successfully separate the texture and structure from a given image while preventing artifacts such as staircase effects present in traditional filter based approaches. The experiment results prove the effectiveness of our approach in image decomposition. We also show the usefulness of our approach by presenting successful applications in structure–texture–noise decomposition, detail enhancement and image abstraction.