Towards Better Visual Comparison for Authenticating Ancient Chinese Paintings
Human Computer Collaboration, Image Analysis, Visual Analysis
Project Overview
This project addresses the challenges faced by experts in authenticating ancient Chinese paintings ("Mù Jiàn," or connoisseurship) due to low efficiency and lack of effective visual aids. We propose an interactive visual analysis method for computer-aided authentication and designed the system CAPAT (Computer-Aided Painting Analysis Tool). The system offers a visual comparison approach for high-definition digital images of traditional paintings, significantly improving the efficiency and accuracy of expert assessment.
Core Value Proposition:
Transforms the traditional, experience-dependent "Mù Jiàn" process into an efficient, interactive human-computer collaborative analysis workflow.
Systematically maps image processing algorithms to the unique artistic characteristics of Chinese ancient paintings (e.g., brushwork, ink application, composition, seals) for visualization.
Research Background & Challenges
With the growth of digital humanities, high-definition image data of ancient paintings is becoming increasingly available. However, existing computer analysis methods struggle with the unique "Xieyi" (freehand) style and complex artistic features of Chinese ancient paintings.
Low Efficiency: Experts often need to manually and empirically retrieve and compare numerous works to analyze a single painter's style, a process that is time-consuming and laborious.
Technology Gap: The complexity of semantic information in artworks and the opaqueness of computer algorithms create a burden for experts attempting to use computer-aided authentication tools.
Single Dimension: Current image analysis methods often analyze paintings from only one or a few dimensions, lacking the capability for multi-dimensional comparison and exploration.
Core Methodology
To address these challenges, we designed an image comparison analysis framework for the visual authentication of ancient Chinese paintings, consisting of three core modules: Human-Computer Collaborative Segmentation, Feature Extraction and Computation, and Comparison Analysis.
Human-Computer Collaborative Segmentation:
Innovation: Integrates deep learning models (such as SAM) with traditional interaction strategies (e.g., rectangular selection, scribbling paths) to achieve efficient and precise segmentation and selection of complex local elements (e.g., figures, mountains, inscriptions) within the painting.
Multi-Dimensional Image Feature Extraction and Computation:
Technology Integration: Combines ResNet feature vectors with the results of traditional image processing algorithms (e.g., SIFT, HOG, edge detection) using weighted fusion to construct a comprehensive feature vector that includes both global information and local details.
Similarity Enhancement: Introduces the Triplet Loss function, commonly used in deep learning, to fine-tune the feature vectors, thereby enhancing the similarity representation of images with the same style or brushwork in the embedding space.
Impact
Based on the methodology above, we designed the visual analysis system CAPAT. It consists of a Segmentation Operation View, a Timeline Exploration View, and a Comparison Analysis View, significantly improving expert efficiency and accuracy in ancient painting research.
Timeline Exploration View:
Adjusted the traditional RadViz chart into a fan-shaped layout, incorporating radial weighting to fuse image similarity and category labels.
Integrated a timeline overview, helping experts intuitively understand the evolution trend of artistic features during an artist's career while focusing on similar images.
Comparison Analysis View:
Offers two modes: overlay comparison and parallel comparison. The overlay mode automatically aligns local images and highlights difference regions, which is especially effective for precise authentication of seals or brushstrokes.
Connects analysis dimensions (e.g., "Brushwork," "Ink Application") to underlying computer algorithms, allowing experts to select corresponding analysis functions directly without needing to understand complex technical details.
User studies and case analyses (e.g., analyzing the wear changes in Shitao’s seals and variations in his brushwork characteristics) validated the effectiveness of the analysis method and visual design. The system significantly improved expert analysis efficiency and introduced a new computer-image-analysis perspective to the authentication of ancient paintings.
research
other
