An efficient seed points selection approach in dominant color descriptors (DCD)

Abstract

The content-based image retrieval (CBIR) system accepts the input in the form of images and retrieves the relevant images from the database. The CBIR system automatically extracts the prominent key information from the image involved in the retrieval task. The color is one of the key information of the image and it is represented by dominant color descriptors (DCD). Here, similar colors get clustered and the mean value of each cluster represents the dominant color. The random number of unstable cluster formation in DCD alleviates the CBIR system performance. The proposed work has minimized the drawback of DCD by introducing seed points selection based on the mean, maximum and minimum value of the color pixels present in the image. Moreover, this work suggests the optimal cluster number by validating the different combinations of the proposed stable dominant color clusters. The retrieval precision of the proposed CBIR has improved since this work gives equal weight for both the dominant color and its occurrence probability in distance metric calculation. Finally, four standard datasets namely Wang’s, Corel-10k, OT-scene, and Oxford flower are considered for evaluation, and it gives more number of relevant images compared to the state-of-the-art dominant color feature extraction techniques used on these datasets.