ChromaFusionNet (CFNet): natural fusion of fine-grained color editing

Abstract

The goal of digital image enhancement is to create visually appealing images that reflect human perception accurately. While global enhancements improve the overall look, precise, localized color adjustments are challenging yet crucial for enhancing visual richness. Existing methods struggle with maintaining consistency, particularly at boundaries. ChromaFusionNet (CFNet) introduces a method by considering color fusion as an image color inpainting issue, using Vision Transformer architecture for comprehensive context capture and high-quality output. It ensures smooth color transitions and boundary preservation. Studies on ImageNet and COCO datasets confirm CFNet’s efficiency in achieving color harmony and fidelity. Its utility is further supported by robustness tests and user feedback, representing a step forward in precise color editing.