Modeling visible differences: The computational observer model

We describe a new approach to the design of visible difference metrics that is based on calculating the information available at different points in the visual pathway and determining whether an ideal linear classifier can use this information to reliably predict the difference between two stimuli. This approach, referred to as the computational observer model, establishes a near upper bound on performance by assessing whether the information necessary to make discrimination is present in the visual system. We describe the implementation of a software framework that embodies the computational observer and illustrate how this framework can be used to address specific questions about display design.