High dynamic range (HDR) images are gaining increasing interest and diffusion in the image processing community, because they allow to faithfully reproduce most of the information present in real world scenes; however, they can not be displayed on a standard monitor unless some processing is performed. Global operators map each pixel individually with the same nonlinear function; local operators, instead, take into account also the pixel neighborhood. Advanced local operators can achieve a better quality; on the other hand, global operators typically have a much lower computational cost and are easier to tune, and are therefore attractive for real-time processing. Global operators are also intrinsically free from a class of artifacts, such as halos or noise amplification, which are often found in local operators. Finally, a global mapping is often a building block of more advanced methods. In this paper we present some psychophysical experiments for the evaluation of a simple global operator. We consider a modified logarithmic mapping, which was shown to have a good performance and depends on a single parameter. In our experiments, several users adjusted the parameter manually in the same viewing conditions for a set of test images. We verified if different users agree on the same or reasonably close values of the parameter for the same test image. We also investigated how the user-adjusted parameter is related to the statistical properties of the image, and found that a percentile provides the best fit. This result allows to compute the parameter automatically for most images.
Psychophysical Evaluation and Tuning of a Simple Tone Mapping Operator
GUARNIERI, GABRIELE;CARRATO, SERGIO;RAMPONI, GIOVANNI
2007-01-01
Abstract
High dynamic range (HDR) images are gaining increasing interest and diffusion in the image processing community, because they allow to faithfully reproduce most of the information present in real world scenes; however, they can not be displayed on a standard monitor unless some processing is performed. Global operators map each pixel individually with the same nonlinear function; local operators, instead, take into account also the pixel neighborhood. Advanced local operators can achieve a better quality; on the other hand, global operators typically have a much lower computational cost and are easier to tune, and are therefore attractive for real-time processing. Global operators are also intrinsically free from a class of artifacts, such as halos or noise amplification, which are often found in local operators. Finally, a global mapping is often a building block of more advanced methods. In this paper we present some psychophysical experiments for the evaluation of a simple global operator. We consider a modified logarithmic mapping, which was shown to have a good performance and depends on a single parameter. In our experiments, several users adjusted the parameter manually in the same viewing conditions for a set of test images. We verified if different users agree on the same or reasonably close values of the parameter for the same test image. We also investigated how the user-adjusted parameter is related to the statistical properties of the image, and found that a percentile provides the best fit. This result allows to compute the parameter automatically for most images.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.