Abstract
The perceptual judgements underlying object recognition are thought to occur by matching visual stimuli against some internal template. However, the visualization of such templates remains difficult given their inaccessible nature. Nonetheless, a paradigm known as psychophysical reverse correlation (PRC) has been utilized to render visible the internal templates involved in such perceptual judgements. On each trial of a PRC experiment, a base image (on which a perceptual judgment is to be performed) is presented in the presence of randomly generated noise. The resulting increase in perceptual ambiguity affects the perception of the base image and subsequent perceptual judgements. Next, behavioral responses are used to combine the noise fields so as to highlight the pixels most diagnostic of participants’ behavior. The resulting image, known as a classification image (CI), can be thought of as a visual representation of the underlying internal template. Although powerful, this method often requires many thousands of trials (Brinkman, Todorov & Dotsch, 2017), which greatly reduces its practicality. To address this, several papers have explored new techniques for CI generation and analysis. However, the existing literature offers few suggestions regarding noise selection. We used simulations to generate CIs for base images consisting of oriented gratings along with different noise types (white, gaussian, & sinusoid). Our results demonstrate that noise type can interact with the base image to significantly impact perception. Specifically, for base images (sine-wave gratings) of lower spatial frequency, all three noise types produced similar results. However, for higher spatial frequency, the white & gaussian noise outperformed the sinusoid noise with respect to CI quality. These findings suggest that a noise type properly tailored to the base image can reduce the number of trials required for CI generation thereby overcoming a key practical limitation of the PRC method.