Content-based image retrieval: Colorfulness and Depth visual perception quantification

Solène Vilfroy, Thierry Urruty, Philippe Carré, Lionel Bombrun, Arnaud Bour

Abstract: This study focuses specifically on content-based image retrieval. The objective being to characterize the images according to their content, in the form of an index, a score which will be used for the manipulation of the images.We propose two visual perception quantification methods, one for colorfulness and one for depth. In terms of quantifying an impression, it was necessary to set up objective and rigorous criteria reflecting a sometimes subjective concept.Based on saturation, lightness and color disparity, we propose a framework that assigns to image an index between 0 and 100 to convey the impression that the image is very colorful or not. While in the case of depth, the estimate of a depth map will serve as a starting point. We estimate that the impression of depth is linked to the spread of the elements that make up the image from the farthest to the closest.