<title>Evaluation of human vision models for predicting human observer performance</title>

A Perceptually Relevant Channelized Joint Observer (PCJO) for the Detection-Localization of Parametric Signals

“…On the assumption that radiologists know what a healthy brain image should be like and refer to it in their mind during the diagnostic process, we take an image acquired from a healthy person as Reference image and the same image with simulated signals (modeled by Eq. (3)) as Distorted image, as Jackson did in [26]. Thus, the VDP actually predicts the probability that the signal is visible in radiologists' perceptual domain.…”

“…The HVS models used in the evaluation of medical image quality [10]- [12], [26]- [28] belong to the perceptual difference model (PDM) [29]- [33] focusing on the spatial contrast detection. They are especially efficient for the detection of near-visibility-threshold distortion, suitable for the diagnostic task performance evaluation where the detection targets are not too conspicuous.…”

Grading Diabetic Retinopathy Severity From Compressed Digital Retinal Images Compared With Uncompressed Images and Film

“…Lower JNDs represent higher fidelity. 19 The JND tools have been reported to be a better predictor of human observer performance than parameters such as peak signal-tonoise ratio or mean-squared error. 34,35 The JNDmetrix visual system model has predicted and correlated highly with radiologists' performance in using softcopy display monitors with different phosphors 13 and on-axis viewing liquid crystal display displays compared with CRT viewing.…”

“…A JND value of 1.0 is defined as a 75% probability that a human observer will notice a difference between the 2 images. 19 Authors (H.L. and J.F.F.-A.)…”

<title>ROC theory under the exponential assumption to analyze visual detection experimental data</title>

“…Since long ago, ROC theory has been applied to analyze signal detectability and discrimination data [1-3, to assess image quality [4,5], ], to measure performance of human observers [6][7][8], and to evaluate diagnostic systems [9][10][11]. An interesting characteristic of ROC theory is that it provides a detectability (or performance) index (d') that is independent of the circumstances that "force" the observer to set his (her) criterion at a particular point of the variable decision axis.…”