Work

Optimizing AI-generated image metadata with hybrid color analysis and semantic keyword structuring

Year: 2025

Type: article

Abstract: Effective metadata optimization is crucial for improving the retrieval and classification of AI-generated images, with color playing a significant role in visual perception and searchability. This study proposes a hybrid metadata optimization framework integrating color-based feature extraction (K-Means clustering and Saliency Detection) with semantic keyword structuring to enhance metadata accuracy and keyword relevance. By combining global color distributions, subject-focused visual attributes, and AI-driven contextual analysis, the proposed method ensures structured and comprehensive image content representation. The methodology comprises three primary stages: (1) Hybrid Color Extraction, (2) AI-based Keyword Generation, and (3) Structured Keyword Optimization. The hybrid extraction process initially employs K-Means clustering to identify globally dominant colors, followed by Saliency Detection to highlight subject-specific hues. Extracted colors are then mapped to descriptive keywords, complemented by context-based keywords generated through an AI captioning model. The final keyword optimization phase systematically categorizes these terms into subject-based, color-based, and descriptive-emotional keywords. The effectiveness of the proposed approach is quantitatively evaluated using several performance metrics, including precision, recall, F1-score, false positive rate, top-10 retrieval accuracy, cosine similarity, Jaccard similarity, and coverage score. Experimental results demonstrate that the proposed framework achieves a precision of 92.10%, significantly enhancing retrieval accuracy and keyword structuring compared to conventional approaches and outperforming state-of-the-art baseline methods, including the Google Cloud Vision API. This research provides a scalable and efficient metadata enrichment solution applicable to digital libraries, image search engines, and content management systems, ensuring accurate, structured, and contextually relevant metadata for effective image retrieval.

Source: Egyptian Informatics Journal

DOI: 10.1016/j.eij.2025.100775

Author: Akara Thammastitkul

Institution: Burapha University

Language: English

Topic: Image Retrieval and Classification Techniques

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $1,200

Funder: Burapha University