Image Pattern Analysis Using Hybrid Feature Descriptors for Content-Based Image Retrieval

Content-Based Image Retrieval (CBIR) has emerged as a vital technique for efficient image indexing and retrieval in large-scale databases. Traditional approaches often rely on handcrafted features such as color, texture, and shape, which are limited in capturing high-level semantic information. On the other hand, deep learning-based descriptors have shown significant improvement in extracting abstract features but often require large datasets and computational resources. This paper proposes a hybrid CBIR framework that fuses deep learning-based RGB image features with texture-based descriptors such as Local Binary Patterns (LBP), Center Symmetric LBP (CSLBP), and Local Directional Patterns (LDP). The hybrid descriptors are extracted using dual convolutional neural network (CNN) encoders, each trained on RGB and texture images independently. Additionally, a weighted feature fusion technique in the CIE Lab* color space is presented to emphasize salient regions of images, enhancing retrieval precision. The system is evaluated on standard benchmark datasets including Corel-1K, Caltech-256, and 102Flower, and achieves superior performance in terms of precision, recall, and F1-score. The experimental results demonstrate that hybrid feature fusion significantly improves retrieval accuracy compared to traditional and single-feature descriptors.