Effective features in content-based image retrieval from a combination of low-level features and deep Boltzmann machine

Image retrieval is a convenient way to browse and search for a set of similar images. The main challenge of Content-based Image Retrieval (CBIR) systems is to extract the appropriate feature vector for image description. In this research, a content-based image retrieval model focusing on extracting effective features is introduced. The introduced feature vector is a combination of low-level and mid-level Image Features. The extraction of low-level features of the image, including color, shape, and texture, was performed using auto-correlogram, Gabor wavelet transform, and multi-level fractal dimension analysis. The mid-level features of the image are also extracted through the use of the Deep Boltzmann Machine as well as by learning the low-level features of the image and the relationships between them. The resulting feature vector of Image retrieval based on the combination of low-level features and deep Boltzmann machine (LB-CBIR) is adjusted with the Corel 1 K dataset, and the performance of the proposed model is measured on the Corel 1 K-illumination, Corel 1 K-Scale, Corel 5 K, Corel 10 K, Oxford buildings and Caltech-256 dataset. The best-evaluated results on the mentioned datasets have been reported with the average precision criterion as 99.4% for Corel 1 K dataset, 94.2% for Corel 1 l-Scale, 82.05 for Corel 1 K-illumination, 98.2% for Corel 5 K dataset, 90.2% for Corel 10 K dataset, 64.1% for Oxford buildings, 32.12% for Caltech-256 dataset. Explainability of the feature vector and the value of the extracted features in the proposed model are also interpreted by calculating the Shapley value.