An Effective Fusion of a Color and Texture Descriptor for an Image Retrieval System: An Exploratory Analysis

Abstract

Content-based image retrieval (CBIR) is one of the most prominent systems by which the desired and relevant images are recovered from a massive database by using the basic image features like color, shape, texture, spatial information, and edge. In this paper, an experimental analysis is being done to determine the most efficient combination using color and texture descriptors. Here, color moment (CM) is used for color feature extraction and is used individually in a combination with different texture descriptors, namely discrete wavelet transform (DWT), Gabor transform, Curvelet transform, graylevel co-occurrence matrix (GLCM), and local binary pattern (LBP). These color and texture features can be combined using two different levels of the system: feature-level fusion and score-level fusion. Both the feature-level and score-level combination techniques are used in this paper. The results of this analytical experimentation depict that the framework of CM, and GLCM attains the highest results among all the other combinations using feature-level fusion technique on a benchmark CBIR dataset, particularly WANG. Precision, recall, f-score are some of the evaluation parameters which are utilized in this paper to measure the effectiveness of the fused descriptors.