New Dominant Color Descriptor Features Based on Weighting of More Informative Pixels using Suitable Masks for Content-Based Image Retrieval

Abstract

Content-based image retrieval (CBIR) is a process of retrieving images based on their content in a dataset automatically. CBIR is a common solution to search images similar to a desired image among all images in dataset. To do this, many methods have been developed to extract images features. Here, a new Dominant Color Descriptor (DCD) method is proposed to improve CBIR accuracy. In the first step, Canny edges of images are extracted. In the next step, edges are widened by employing morphological operations. Finally, pixels that are not at the edges are weighted less than the pixels which are located at edges. Indeed, pixels in regions with low color variations are less weighted and more informative pixels are more weighted in providing DCD features. To show the effectiveness of the proposed method, experiments are performed on three datasets Corel-1k, Corel-10k and Caltech256. Results demonstrate that the proposed method outperforms competitive methods.