Time to wave good-bye to phase scrambling: creating controlled scrambled images using diffeomorphic transformations

Area of Science:

• Neuroscience
• Computer Vision
• Cognitive Science

Background:

• Disentangling neural mechanisms for object recognition requires control stimuli with preserved perceptual properties but lacking recognizability.
• Conventional scrambling methods (phase, box, texture) distort essential visual features, making them unsuitable controls.

Purpose of the Study:

• To develop and validate a novel image scrambling technique preserving basic visual properties while eliminating object recognition.
• To evaluate the efficacy of this new method against traditional scrambling techniques in computational models of visual processing.

Main Methods:

• A diffeomorphic transformation was employed to create "diffeomorphed" scrambled images.
• Perceptual ratings were collected to determine minimal scrambling for loss of recognition.
• The HMAX computational model simulated neural activity in early visual layers (S1, C1, S2) for intact and scrambled images.

Main Results:

• "Diffeomorphed" images were computationally indistinguishable from intact images in early visual processing layers.
• Traditional scrambling methods (phase, box, texture) produced significantly different neural activity patterns compared to intact images.
• Perceptual ratings confirmed that "diffeomorphed" images effectively removed recognizability while retaining visual characteristics.

Conclusions:

• "Diffeomorphed" images serve as superior control stimuli for neuroimaging research.
• This method aids in isolating neural representations of perceptual versus semantic object properties.
• The findings advance the development of more accurate experimental designs in visual neuroscience.