David Serrano-Lozano,Luis Herranz,Michael S. Brown,Javier Vazquez-Corral
Abstract:A popular method for enhancing images involves learning the style of a professional photo editor using pairs of training images comprised of the original input with the editor-enhanced version. When manipulating images, many editing tools offer a feature that allows the user to manipulate a limited selection of familiar colors. Editing by color name allows easy adjustment of elements like the "blue" of the sky or the "green" of trees. Inspired by this approach to color manipulation, we propose NamedCurves, a learning-based image enhancement technique that separates the image into a small set of named colors. Our method learns to globally adjust the image for each specific named color via tone curves and then combines the images using an attention-based fusion mechanism to mimic spatial editing. We demonstrate the effectiveness of our method against several competing methods on the well-known Adobe 5K dataset and the PPR10K dataset, showing notable improvements.
What problem does this paper attempt to address?
### What problems does this paper attempt to solve?
This paper aims to solve several key challenges in image enhancement, especially in imitating the editing styles of professional photographers or colorists. Specifically:
1. **Complexity of manual image adjustment**: Although camera technology has made remarkable progress, many amateur and professional photographers still need to perform image enhancement after shooting to improve image quality. However, manual enhancement can be very difficult for those lacking professional knowledge, time, and a good sense of aesthetics.
2. **Automatically learning professional editing styles**: To address the complexity of manual adjustment, researchers have proposed a deep - learning - based method that uses pairs of training images (original input and edited versions) to learn the styles of professional photo editing. These methods utilize data - driven techniques, enabling the model to imitate the editing methods of experts.
3. **Image decomposition and adjustment based on color naming**: Many image - editing tools allow users to make adjustments based on a fixed set of colors (such as red, green, yellow, orange, blue, purple, etc.). This editing method based on color names enables users to easily adjust elements such as the "blue" of the sky or the "green" of trees. Inspired by this, the authors proposed the **NamedCurves** method, which decomposes an image into a set of named colors and enhances the image by learning the global adjustment of each specific named color.
4. **Fusing local and global editing effects**: To simulate the spatial editing effects of experts, the authors introduced an attention mechanism to fuse the images adjusted by different color curves. This method not only achieves global color adjustment but also can simulate local editing effects, thus being closer to the results of professional editing.
### Specific objectives
- **Improve image quality**: Automatically generate high - quality enhanced images by learning the styles of professional editing.
- **Simplify the image - editing process**: Enable users to obtain professional image - enhancement effects without professional knowledge.
- **Improve existing methods**: Verify and demonstrate the superiority of this method over other existing methods on multiple public datasets, especially in terms of color fidelity and visual effects.
### Main contributions
- Proposed **NamedCurves**, a learning - based image - enhancement technique based on color - naming decomposition.
- Used smooth Bezier curves to estimate the hue curves of each named color.
- Introduced an attention - mechanism - based fusion scheme to combine the images adjusted by different color curves to simulate spatial editing effects.
- The experimental results on the MIT - Adobe - 5K and PPR10K datasets show that this method is superior to existing methods in terms of indicators such as PSNR and ∆E.
Through these improvements, the authors hope that NamedCurves can provide users with a more intelligent and efficient image - enhancement tool.