For the development of objective visual quality metrics and visual attention models, a ground truth is typically needed on which to design and validate the models on. In case of quality metrics, mean opinion scores (MOS) obtained in subjective quality experiments are typically utilised. On the other hand, gaze patterns from eye tracking experiments often support the design and validation of visual attention models. For a better comparison of the objective methods between laboratories worldwide it is desirable to have publicly available quality and eye tracking databases. The purpose of this website is to make such databases available to the research community. Until now, we have conducted several subjective experiments in three different laboratories involving numerous participants. The outcomes of these experiments serve to develop and validate objective quality metrics and visual attention models and further provide valuable insight to gain a better understanding of human perception of visual scenes.
We created three databases that we make available to the research community. Each of the databases contains subjective data that has been obtained in experiments involving a number of human observers. Detailed information about each database can be obtained by following the respective link below.
Visual Attention for Image Quality (VAIQ) database
The VAIQ database contains gaze patterns and saliency maps for the reference images of three well known image quality databases (IVC, LIVE, MICT). These gaze patterns were obtained from an eye tracking experiment involving 15 human observers.
Wireless Imaging Quality (WIQ) database
The WIQ database contains test images that were distorted by a wireless communication link. In two extensive subjective experiments, these images were presented to 60 human observers who were asked to rate their quality. The subjective scores obtained from the experiments are also provided in the database.
Region-of-Interest (ROI) database
This database contains region-of-interest (ROI) selections for the reference images used in the WIQ database (see above). The ROI were manually selected in an experiment involving 30 participants.