The human perceptual decision of the quality of images is a mechanism that is not yet fully understood. In the special case of human faces, quality represents a critical element for mechanisms that include the efficiency with which human beings comply with facial recognition, among many other mechanisms and applications of them. On the other hand, automatic systems for evaluating the quality of images, facial or other, in particular systems based on artificial neural networks (deep learning), make the computational evaluation of the quality of an image through quantitative metrics. An example is facial recognition systems. Another example is related to systems for protecting the integrity of face images for the issuance and validation of documents such as passports or citizen cards.
What has been found in recent studies is that there is a large disparity between human perceptual decision (based on perception mechanisms) and quantitative metrics, generally based on the pixels of images, but with little relation to true perception.
This project aims to help understanding the differences between quantitative metrics and the human perceptual decision of the quality of images, in particular images of faces.
To study this issue, several image distortion scenarios can be considered, particularly steganography, distortion of JPEG compression and other distortions.
You are about to participate in a research study. The data you provide during this study (gender, age, degree, country of origin) will be used for research purposes for a MSc thesis. Your participation is voluntary, and all information you provide will be kept confidential.
In this test, you will evaluate the visual quality of a series of individual images. For each image, you will assign a score based on its perceived quality using a predefined scale. Please read the following instructions carefully before starting.