Psychophysical Experimental Methods

Despite being an archaic adage, “a picture is worth a thousand words” really captures the complexities that exist within a single photo and how it can emotionally impact an individual. Pictures are amazing in this way, since they can fully capture the feelings of a particular moment and convey them to anyone who views them. ISO 20462-1:2005 - Photography - Psychophysical Experimental Methods for Estimating Image Quality - Part 1: Overview of Psychophysical Elements provides the background necessary to quantify a subjective understanding of this for any particular image.

Psychophysics is a branch of the study of perception that looks at the relations between observed stimuli and responses and reasons for those relations. It is based on the assumption that the human perceptual system is a measuring system yielding results that can be analyzed. In the past, psychophysics in application to image quality has mostly been measured to ambiguous results. However, more recent calibrated psychometric methods of developing comprehensive models of imaging system quality have great value in actually capturing these elements of an image.

One of the more common psychophysical methods of determining image quality is the paired comparison method, in which an observer is given the choice between two simultaneously presented stimuli, and his or her reaction is considered to determine which stimuli has greater quality. ISO 20462-1:2005 establishes that the main units of stimulus differentiation from psychophysical experimentation are that of just noticeable difference (JND), which is considered to be the stimulus difference that would lead to a 75:25 proportion of responses in a paired comparison task. The standard criticizes this method, stating that it requires a large amount of assessments to analyze the results.

The other two parts of this standard establish additional methods that can successfully capture a subjective visual assessment of a photo. ISO 20462-2:2005: Photography - Psychophysical experimental methods for estimating image quality - Part 2 describes the triplet comparison method. This methodology manages the problem with the paired comparison method of requiring too many assessments by making use of an initial step, called the category step, which reduces the number of samples to an appropriate level by organizing them into three categories, such as “favorable”, “acceptable”, and “unfavorable”. After this is the triplet comparison step, in which samples are organized into groups of three that are compared to each other, achieving high assessment accuracy while keeping experimental scale realistic. This second stage should allow the observer to place the samples into five categories: favorable, acceptable, just acceptable, unacceptable, and poor. With this data, the observations can be used for statistical analysis to obtain an interval scale, which can then be converted into JNDs.

The third method, which is detailed in ISO 20462-3:2012: Photography - Psychophysical experimental methods for estimating image quality - Part 3, also involves the visual comparison of two or more image samples, except it makes use of reference stimuli to compare to the sample. The quality ruler method involves the use of univariate series of reference stimuli depicting the same scene and having known stimulus differences between them expressed in JNDs of quality. Alongside the sample, the reference stimuli are presented to the observer, who must determine which is the closest in comparison of the sample. This determines the JND amount of the sample, since there is already a known quantity attached to the reference. The primary advantage of this procedure is that the ratings given by an observer are converted into JND in real time, as opposed to the triplet comparison and paired comparison methods, which require the use of calculations to follow up the observations.

This figure from ISO 20462-3:2012 depicts the apparatus from the quality ruler method.

One limitation of the quality ruler method is that the reference stimuli cannot be less than one JND. While the triplet comparison method and the quality ruler method can be used individually, they are complementary to each other and can be used together to achieve well-calibrated results.
Share on Google Plus


Post a Comment