Visual Attention Modeled with Luminance Only: from Eye-tracking Data to Computational Models

Research on image quality assessment has shown the potential performance enhancement of adding visual attention in objective metrics. However, the use of attentionbased metrics in real-time applications is mainly limited by the complexity of modeling visual attention. Since most of the existing objective metrics are based on the luminance component of images only, we investigate whether also saliency can be modeled only with the luminance component. An eye-tracking experiment was carried out to subjectively assess the contribution of color in saliency. Additionally, two attention models well-known in literature were evaluated on predicting our eye-tracking data. Our experimental results show that human saliency is insensitive to color, but that this is not predicted by the attention models existing in literature.