Dominant Colors as Image Content Descriptors: A Study with Users

Abstract

Image content are typically described using low level features such as color, texture, shape, or a combination of the previous. A particular use of color is the identification of the dominant colors in images to describe its content, for image retrieval, for instance. In this paper, we present a study with users to verify if the dominant colors can be used as image content descriptors. From the study we identified the dominant and the search colors users associated to a set of images. We supplemented this information with gaze coordinates, collected with an affordable eye tracker, to register the regions at which people looked while identifying colors in the images. The analysis of the data revealed that users used a small set of color names, and that the colors used for searching were similar to those considered dominant, validating the use of dominant colors as image descriptors. As a result of the study, we make available a dataset of 100 images annotated with their dominant colors, the colors that users would use to search for them, and the areas where they looked while identifying both types of colors.