Answers are grounded in this paper and cite their sources.
This paper proposes the Complex Network Image Transform (CNT) to enhance texture characterization and classification. Our approach leverages the capability of complex networks to effectively represent and analyze intricate phenomena and patterns by modeling the image as a complex network that encodes the detailed relationships among pixels. The topological properties of the resulting network are then employed to generate a set of images that emphasize complex texture patterns, capturing both local and global characteristics through a predefined modeling parameter. The key insight is that these transformed images can serve as enriched inputs for texture analysis and classification tasks. Thus, we evaluate CNT in two distinct scenarios: (i) as a preprocessing step to enhance handcrafted texture descriptors, and (ii) as a data augmentation strategy to improve the training of deep convolutional neural networks. Experiments conducted on five texture datasets show that CNT-generated images are rich in discriminative information. Classification results consistently outperform several baselines, indicating that CNT enhances feature extraction and boosts model generalization. These findings indicate the potential of CNT as a useful tool for texture analysis, and suggest that network-based representations may be worth exploring in related image analysis tasks.
No takes yet. Share an insight, caveat, or question.
Zerati et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: