22 December 2015
Despite the obvious difference between a chihuahua and a doberman, the human brain effortlessly categorises them both as dogs, a feat that is thus far beyond the abilities of artificial intelligence.
Previous research has established that the brain can recognise and categorise objects extremely rapidly, however the way this process occurs is still largely unknown. Researchers from Monash University have pioneered a new image modulation technique known as semantic wavelet-induced frequency-tagging (SWIFT) to further test how images are processed.
This work, by Associate Professor Naotsugu Tsuchiya and Dr Roger Koenig-Robert from the School of Psychological Sciences (affiliates of Australian Research Council Centre for Integrative Brain Functions), identifies a way to visually stimulate the brain to isolate neural activity responsible for categorising objects. This is not an easy task at all, because areas in the visual cortex supporting these category representations are simultaneously active with areas representing low-level visual features such as lines, forms and contrast.
Categorisation of objects is believed to emerge gradually, in a hierarchical manner. For example, simple visual features such as lines, contrast and colour are thought to be represented at early stages. For more complex representations, these features are combined, giving rise to abstract categorisation (such as cars, faces and animals), and neurons in high-level areas are believed to be responsible.
By isolating the areas representing abstract, object categories in the brain, the research has allowed greater understanding of how humans effortlessly categorise objects, despite massively different appearances.
Associate Professor Tsuchiya said the discovery could prove useful for manipulating images so that they communicate information in a subliminal, non-conscious manner.
“SWIFT can degrade arbitrary natural stimuli in a subtle manner. This technique may find its own application as a way to reveal and/or hide specific aspects of images. Potentially, SWIFT could be used to probe robustness and flexibility of artificial visual systems, such as those used in security,” Associate Professor Tsuchiya said.
Using functional magnetic resonance imaging (fMRI), the study determined that in faces and scenes modulated with SWIFT, there was a sustained and constant brain response in early visual areas, and periodical responses in the higher level (category-selective) areas.
The cutting-edge research has been published in the prestigious PLOS ONE journal and can be viewed here.