When images that differ significantly in creation, display, and content are being compared and intervalscaling becomes necessary, interpretation of the results become more and more complex as the number of observers is increased. Ratio-scaling provides a means for correlating multiple observations and multiple data sources. Observers are asked to assign a numerical value to perceived image quality (either in total or with regard to a specified characteristic such as noise, resolution, color rendition, etc.). They are also asked to identify numerical values for the best possible image, and the completely unacceptable image. Each is allowed to choose a numerical scale with which the observer feels most comfortable. The relationship between the value for the test image and the two extremes provides a useful ratio.

Analyses involving comparisons among observers, comparisons with other systems, correlation of results obtained over periods of time, etc., are made by normalizing each observer’s scale (for example, best possible = 100, completely unacceptable = 0).