Saliency Models as Gamut-Mapping Artifact Detectors

When an image is reproduced with a device different artifacts can occur. These artifacts, if detectable by observers, will reduce the quality of the image. If these artifacts occur in salient regions (regions of interest) or if the artifacts introduce salient regions they contribute to reduce the quality of the reproduction. In this paper we propose a novel method for the detection of artifacts based on saliency models. The method is evaluated against a set of gamut mapped images containing the most common artifacts, which have been marked by a group of color experts. The results have shown that the proposed metrics are promising to detect the artifacts through the reproduction. Introduction For centuries researchers have been trying to achieve accurate reproductions of images. Technological advancements have taken us closer and closer to this goal, but still it has not been reached. When we reproduce an image several issues, such as the limited number of colors a system can reproduce, contribute to the quality of the reproduction. One of the attributes which influences image quality greatly is artifacts. We will focus on the reproduction of images in a print work flow, more specifically the gamut mapping process. When we gamut map an image several artifacts may occur, such as contouring, loss of details, and halos. These artifacts contribute to reducing the quality of an image if detectable [1–3]. One way of detecting them is a visual inspection of the image. However, this might be time consuming and resource demanding. Because of this objective evaluation methods have been proposed. There have been proposed several objective methods, commonly referred to as image quality metrics [4]. These traditional metrics such as MSE and PSNR, are not credible for evaluating the image quality when the artifacts are introduced. This is due to the fact that the metrics do not take the characteristics of the Human Visual System (HVS) into account. The goal of this paper is to detect gamut mapping artifacts. In order to achieve this goal we propose a novel method based on saliency. We limit in this paper our method to detection of loss of details and contouring in shadow regions, where the artifacts are usually most perceivable and detectable. This paper is organized as follows: first we present state of the art, then a section on the applicability of saliency maps to gamut mapping artifacts, before we introduce a new method for using saliency models to detect gamut mapping artifacts. Experimental results are shown before we conclude. State of the art Artifacts The artifacts found in an image from a reproduction system can severely degrade the quality of an image, and is considered as one of the major attributes contributing to influencing image quality [4, 5]. Several researchers have stated the importance of artifacts for color printing; Bonnier et al. [6] found that artifacts strongly influence the quality of gamut mapped images. In an experiment by Hardeberg et al. [7] observers stated loss of shadow details as the most important criteria for the evaluation of gamut mapped images. Pedersen and Hardeberg [4] evaluated a color work flow, and found that observers looked for artifacts in almost 40 percent of the images they evaluated. Evaluating images in a psychovisual test is time consuming and requires a lot of observers. Hence researchers aim at automated evaluation with image quality measures. As states [1], metrics of image quality based on detecting artifacts are much more complex than those based on image fidelity. But artifacts strongly affect observers’ preferences. If they occur in the image, measures based on fidelity are usually too weak to model quality of considered images. There has been little recent research on detecting artifacts caused by gamut mapping. Contouring artifacts can be partially identified using methods for finding blocking artifacts in JPEGcompressed images, e.g. [8] Another typical artifact caused by gamut mapping, loss of details, can be probably better handled by structural image quality measures. A recent paper by Zolliker and Simon [9] also discussed removing loss of details artifacts by applying unsharp masking. However this technique cannot be directly used to detect those artifacts. We extend the previous work by integrating the saliency maps with an emphasis in artifact detection.