Image optimization using improved gray-scale quantization for content-based image retrieval

Image quantization is an important process in content-based image retrieval (CBIR) systems. In this study, color-image quantization is used to reduce the number of color bins prior to establishing feature extraction step. We modified and applied Improved Grey-Scale (IGS) procedure to obtain optimized feature representations, by mapping a broad dimension of input pixel distributions of an image to a qualified number of outputs (coded pixel values). Thus, to investigate the impact of IGS method using different parameters on system classifications. In order to determine the number of color intensities that are most appropriate for system performance, a set of image-classification tasks were performed using support vector machine classifier on the Coral-1000 dataset. Accordingly, to evaluate the system accuracy, several metrics are considered in terms of precision, recall, and F1 score of each image category in the dataset. Moreover, we present a comparison of three-color optimization methods that generate compact color dimensions while aiming to preserve or enhance image retrieval performance. The results indicate that our IGS-based CBIR system performs better even when color depths are reduced to three bits in each color component.