Fine-Grained Visual Art Analysis via Yolov9 and K-Means
Bibliographic Data
| ID | 19490486 |
|---|---|
| Authors | Huafei Ma (Naresuan University), Tatiya Theppituck (Naresuan University) |
| Year | 2026 |
| Volume | 7 |
| Issue | 7s |
| Publication date | 2026-05-19 |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | ShodhKosh: Journal of Visual and Performing Arts (JOURNAL) |
| Journal identifiers | ISSN: 2582-7472 • E-ISSN: 2582-7472 |
| Publisher | Granthaalayah Publications and Printers (PUBLISHER • IN) |
| DOI | 10.29121/shodhkosh.v7.i7s.2026.8199 |
| OpenAlex | W7161622964 |
| Language | EN |
| References cited | 24 |
Using the most recent YOLOv9 model and optimized K-means clustering for adaptive feature extraction and objective classification, this research suggests a universal framework for fine-grained visual art analysis. With an emphasis on eliminating subjectivity in conventional visual art analysis, the framework converts qualitative artistic qualities into quantifiable parameters by quantifying 12 fundamental morphological features, such as line curvature (σ=0.85), symmetry index (using Hu moments), and contour regularity (using Fourier descriptors). Tested on a variety of 1,000 high-resolution images (including jade carvings, ceramic patterns, and pieces of murals), the framework partitions samples into five different categories: Normative Geometric, Free-Form Geometric, Natural Bionic, Minimalist Line, and Symbolic Pattern. It achieves automated clustering with an overall accuracy of 92.7% and a silhouette coefficient of 0.91. Three art historians' cross-validation shows 94.8% consistency with expert classifications, greatly exceeding conventional AI techniques (e.g., ResNet50 with 84.2% accuracy on the same dataset). In support of applications in digital archiving, stylistic evolution research, and cultural heritage preservation, this work develops a strong technical instrument for extensive digital analysis of visual arts
Cluster analysis · Feature extraction · Painting · Silhouette · Visualization · Aesthetic Perception and Analysis · Art History and Market Analysis · Generative Adversarial Networks and Image Synthesis
| Highly cited | No |
|---|