LogoRef: A Region-Aware Framework for Reference-Based Logo Sketch Colorization

Reference-based logo colorization aims to generate controllable color schemes while preserving clear contours and consistent colors within graphic regions. However, existing colorization and style transfer methods often suffer from ambiguous color assignment, inconsistent filling, and cross-boundary color interference when applied to logos with compact structures and nested regions. To address these issues, we propose LogoRef, a reference-based logo colorization framework that integrates sketch structure preservation, reference color extraction, and region-aware color assignment. First, we propose a Hierarchical Region Prior (HRP) to extract region structures at different levels from the input line art. Second, we introduce the Semantic Region Style Attention module, which constructs a reference style library and adaptively selects appropriate color statistics for different logo regions, thereby improving intra-region color consistency and reducing color interference between adjacent regions. In addition, we construct JN-Logo-v2, a high-quality logo image dataset that provides richer and more reliable data support for logo colorization. Finally, we propose two task-specific evaluation metrics, Color Fidelity and Boundary Integrity, to evaluate reference color consistency and boundary-aware colorization quality, respectively. Extensive experiments on JN-Logo-v2 and a public anime sketch colorization dataset demonstrate that our method achieves superior performance in both quantitative and qualitative comparisons.