Every year experimental techniques appear or resurface that provide tools and theoretical frameworks with which to study the mechanisms and processes involved in human visual perception. Occasionally, one of these techniques will resonate and trigger a wave of excitement among a group of researchers in the field, and will lead to a large number of studies using the method. This was so in the case of the technique known as classification images which was first presented at the European Conference on Visual Perception (Ahumada, 1996).

The technique has still earlier roots in auditory work (Ahumada & Lovell, 1971), wherein multiple regression analysis was applied to the problem of auditory tone detection in noise, to estimate the contribution of auditory stimulus features to the observer’s decision variable. The regression weights plotted as a function of feature temporal frequency could be called classification plots. The central concept of the technique is the correlation of observer decisions with noisy stimulus features over sets of stimuli. From the correlation of the features with the decisions and the inter-correlations among the features, the investigator can then estimate how the observer is weighting the stimulus features to reach a decision.

In the 1980´s and 1990´s a related technique, reverse correlation, was used to estimate the features of receptive fields of the visual cortex (e.g., Jones & Palmer, 1987; Ohzawa, DeAngelis, & Freeman, 1996; Ringach, Hawken, & Shapley 1997).

Ahumada and Beard applied the technique in visual psychophysics to study vernier acuity tasks (Ahumada, 1996; Beard & Ahumada, 1998). They used image pixel intensity as the stimulus features, so that the stimulus feature weights did form an actual classification image. The study revealed that human observers weighted the visual information of the stimuli differently from the optimum (ideal) observer. However, perhaps more importantly, it provided a first glance at the technique for other scientists, and sparked their imagination about the many questions that might be addressed using the technique. Could it be used to study stereo, illusory contours, letter discrimination and perceptual learning? Could it be used with tasks other than the yes/no task? What would be the best (most efficient) algorithm to estimate the classification images?

Why do classification images appeal to vision scientists? Is it the visual nature of the classification image, or is it its exploratory nature that resembles a sort of archeology of perceptual processes? For whatever reason, soon after 1996 a number of researchers started using the technique to study a variety of problems in human visual perception including letter discrimination (Watson & Rosenholtz, 1997), perceptual learning (Knoblauch, Thomas, & D'Zmura, 1999), illusory contours (Gold, Murray, Bennett, & Sekuler, 2000), foveal versus peripheral vernier performance (Beard & Ahumada, 1999), stereo (Neri, Parker, & Blakemore, 1999), and off-frequency looking in non-white noise (Abbey & Eckstein, 2000).

On the other hand, some researchers have watched with skepticism from the sidelines the development and use of the classification image technique. Can one really know whether an empirical classification image is meaningful or simply noise? Can the experimenter´s interpretation of the classification image be biased so that he/she simply sees confirmatory evidence of a hypothesis? How can one reach any meaningful conclusions when comparing two classification images? What are the underlying assumptions of the technique and how meaningful is the technique when the assumptions are violated? All of these questions should have clear answers for a mature scientific method, and should oblige investigators to use scientific rigor in the use of the technique. These questions have indeed motivated investigators to develop sound procedures to estimate the classification image and to test different hypotheses about the obtained classification images (e.g. Is the classification image statistically significantly different from noise? Is it significantly different from the classification image of the ideal observer?)

The papers in this special issue of the Journal of Vision describe methods to estimate classification images and to test hypotheses about them. The papers also provide examples of the wide variety of topics and questions in visual perception that can be addressed using the technique. As guest editors for this special issue, we hope it will provide a useful starting point for researchers interested in embarking in projects using this novel and exciting technique.