Image color feature extraction methods review

Abstract

Recent research in color feature extraction demonstrates significant advancement from foundational works by Swain and Ballard [2], who established color histograms and histogram intersection methods for image indexing. Huang and colleagues [6] expanded these concepts by introducing color correlograms that incorporate spatial correlations between colors, proving more effective than traditional histogram methods. The standardization efforts of MPEG-7 stimulated development of dominant color descriptors and color layout descriptors, providing standardized solutions for image indexing systems. Mathematical advancements led to sophisticated approaches based on orthogonal moments, including Zernike chromaticity distribution moments for compact color representation and quaternion Zernike moments enabling holistic color image processing through quaternion algebra. The primary objective of this comprehensive review is to systematize modern color feature extraction methods through detailed analysis of scientific literature from leading researchers. This study aims to provide thorough examination of theoretical foundations, practical characteristics, and application features of various approaches to facilitate informed selection of optimal methods based on specific image processing requirements. The main investigation examines twelve fundamental color feature extraction methods categorized into distinct approaches. Global methods include color histograms capturing overall color distribution patterns, histogram intersection measuring color distribution overlap, and color histogram for K-means clustering reducing dimensionality while preserving essential information. Local approaches incorporate spatial information through color correlograms expressing spatial correlation changes with distance, color co-occurrence matrices analyzing spatial relationships using convolution techniques, and color coherence vectors classifying pixels based on coherent region membership. Standardized MPEG-7 descriptors encompass dominant color descriptors providing compact representation and color layout descriptors utilizing discrete cosine transform for spatial encoding. Advanced mathematical methods include color moments employing statistical measures, Zernike chromaticity distribution moments offering rotation invariance, and quaternion Zernike moments enabling comprehensive analysis through quaternion algebraic framework. The systematic analysis concludes that color feature extraction methods effectively divide into global and local categories, each addressing specific application requirements. Global methods provide computational efficiency suitable for basic indexing tasks, while local methods deliver enhanced discriminative capabilities through spatial information integration. Standardized descriptors ensure cross-system compatibility, whereas mathematical moment-based approaches offer superior geometric invariance properties. Optimal method selection requires careful consideration of accuracy requirements, computational constraints, and application-specific characteristics, suggesting future research should focus on integrating complementary approaches to maximize strengths while minimizing limitations.