Comparison Between State-of-the-Art Color Local Binary Pattern-Based Descriptors for Image Retrieval

Abstract

In recent years, color Local Binary Pattern (LBP) based descriptors have garnered substantial attention in computer vision and image analysis. This study presents a comprehensive review of color LBP-based descriptors developed over the past decade, focusing on their performance in image retrieval tasks. The research compares these descriptors based on mean Average Precision (mAP) scores, the dimensionality of their feature vectors, feature extraction time, and retrieval time. From this analysis, the top five descriptors are carefully selected and evaluated across multiple datasets Wang, Corel-5k, and Corel-10K. Among these, Weighted Color Radial Mean Completed Local Binary Pattern (WCRMLBP) emerges as the top-performing descriptor, demonstrating the effectiveness of feature weighting in enhancing color LBP-based descriptor performance. This research emphasizes the progress made in color LBP-based descriptors and their significance in contemporary image analysis and proposes the possibility of additional enhancements through improved feature weighting techniques. It contributes to the continual evolution of image processing and computer vision, particularly in deep learning methods.

Keywords

• Comparative analysis
• color image retrieval review
• color local binary pattern
• deep learning
• statistical methods
• weighted color local binary pattern