Enhancing Content-based Image Retrieval Performance through Optimized Feature Selection

Abstract

Content-based image retrieval systems face significant challenges in maintaining optimal performance when handling large-scale image databases, particularly in balancing retrieval accuracy with computational efficiency. This paper presents a novel hybrid optimization framework that enhances CBIR performance through adaptive feature selection and intelligent fusion strategies. The proposed system implements a multi-level feature extraction approach that combines color, texture, and local descriptors with an optimized weighting mechanism. A dynamic feature selection algorithm minimizes redundancy while preserving discriminative information, significantly improving retrieval accuracy. The proposed system incorporates an efficient indexing structure and cache optimization technique that reduces computational overhead while maintaining performance under increasing database sizes. Experimental results demonstrate the superior performance of the proposed system with a mean Average Precision (mAP) of 0.91 and an average response time of 45.2 ms, representing an 8% improvement in accuracy and 35% reduction in processing time compared to existing methods. The system maintains robust performance under varying query complexities, showing only 7% degradation for complex queries while achieving 86% cache efficiency during extended operations. This comprehensive approach effectively addresses the semantic gap challenge while ensuring computational efficiency, making it suitable for large-scale CBIR applications.