A psychophysical evaluation of techniques for Mooney image generation

TL;DR

This study systematically evaluates how different thresholding techniques and levels of smoothing affect the interpretability of automatically generated Mooney images, revealing that threshold choice is less critical than smoothing parameters.

Contribution

It introduces an automated approach to Mooney image generation and assesses the perceptual effects of various image processing techniques on human interpretability.

Findings

Thresholding technique choice has minimal impact on interpretability.

Smoothing level significantly influences Mooney image clarity.

Automatic generation can replace manual, subjective methods.

Abstract

Mooney images can contribute to our understanding of the processes involved in visual perception, because they allow a dissociation between image content and image understanding. Mooney images are generated by first smoothing and subsequently thresholding an image. In most previous studies this was performed manually, using subjective criteria for generation. This manual process could eventually be avoided by using automatic generation techniques. The field of computer image processing offers numerous techniques for image thresholding, but these are only rarely used to create Mooney images. Furthermore, there is little research on the perceptual effects of smoothing and thresholding. Therefore, in this study we investigated how the choice of different thresholding techniques and amount of smoothing affects the interpretability of Mooney images for human participants. We generated Mooney images using four different thresholding techniques and, in a second experiment, parametrically varied the level of smoothing. Participants identified the concepts shown in Mooney images and rated their interpretability. Although the techniques generate physically-different Mooney images, identification performance and subjective ratings were similar across the different techniques. This indicates that finding the perfect threshold in the process of generating Mooney images is not critical for Mooney image interpretability, at least for globally-applied thresholds. The degree of smoothing applied before thresholding, on the other hand, requires more tuning depending on the noise of the original image and the desired interpretability of the resulting Mooney image. Future work in automatic Mooney image generation should pursue local thresholding techniques, where different thresholds are applied to image regions depending on the local image content.