Summary: New evidence challenges the idea that the visual system classifies objects using only simple, single features.
Source: HSE
Researchers at HSE University in Russia tested whether the human visual system can automatically sort intermixed objects into categories without focused attention. Their straightforward experiment, combined with electroencephalography (EEG) measurements, supports the view that the visual system can perform automatic categorical parsing using combinations of features rather than relying solely on single, simple attributes.
The study was published in Scientific Reports and received support from a Russian Science Foundation grant.
Our visual environment supplies a vast amount of information every moment, yet the brain has limited processing capacity. One common idea is that to avoid overload the visual system compresses incoming information, reducing detail and classifying objects by broad attributes—such as size or orientation—so that later, if needed, a more detailed analysis can follow. The current research asked whether such categorization can occur automatically, without the need for attention, and under what conditions this automatic parsing takes place.
To identify automatic categorization, the researchers relied on the visual mismatch negativity (vMMN) component measured with EEG. vMMN indexes automatic sensory discrimination: it appears when the brain registers a deviant or unexpected visual stimulus among a sequence of standard stimuli, and critically it can be detected even when attention is directed elsewhere.
“We are fascinated by how the visual system can rapidly separate many objects into meaningful groups,” says Vladislav Khvostov, Junior Research Fellow at the HSE Laboratory for Cognitive Research, School of Psychology, and a co‑author of the paper. “For example, when viewing an apple tree we quickly distinguish apples from leaves. Our findings indicate that this quick grouping can arise automatically from sensitivity to differences across objects.”
The experimental task kept participants’ attention tightly focused on a central, asymmetrical cross. Volunteers were instructed to press a button whenever the cross changed orientation, a simple central task that occupied attention and prevented deliberate analysis of surrounding stimuli. At the same time, rows of lines with varying lengths and orientations filled the background. The researchers recorded EEG while participants performed this task to capture brain responses to background changes that the participants were not actively attending to.
In each block of the experiment participants viewed 700 brief displays. Each display appeared for 200 ms and was followed by a 400 ms blank interval. Most displays (standard stimuli) used one consistent mapping of line length and orientation—for example, long lines paired with flat orientation and short lines paired with steep orientation. In 10% of cases a deviant display inverted this mapping: long lines were steep and short lines were flat. Participants were not instructed to attend to these background patterns and typically did not notice the rare changes, but EEG recordings revealed whether their visual systems registered them.
The researchers categorized features as “segmentable” when they had two clear, separate peak values (for example, short vs. long length, or vertical vs. horizontal orientation). Features were labeled “non‑segmentable” if values formed a continuous range without distinct peaks. They tested conditions in which either both attributes were segmentable, only one was segmentable, or neither was.
A clear vMMN response to deviant stimuli appeared when both attributes were segmentable, and also when length alone was segmentable. Because the distributions of length and orientation remained the same within each block, the team concluded that participants’ visual systems were not discriminating stimuli solely by a single feature repeatedly occurring across displays. Instead, the brain detected changes in the conjunctions—specific combinations—of features.
In other words, the visual system not only tracks simple individual attributes but can also automatically parse spatially intermixed objects into categories defined by combinations of features. These results challenge the assumption that unattended visual categorization is limited to single, simple features; the brain can execute a more complex form of grouping even when attention is occupied.
About this visual neuroscience study
Source: HSE
Contact: Liudmila Mezentseva – HSE
Image: The image is in the public domain
Original Research: Open access. “Spatially intermixed objects of different categories are parsed automatically” by Vladislav A. Khvostov, Anton O. Lukashevich & Igor S. Utochkin. Scientific Reports.
Abstract
Spatially intermixed objects of different categories are parsed automatically
The visual system can separate spatially intermixed items into distinct categorical groups (for example, berries versus leaves) by using the shape of feature distributions: if a feature’s distribution shows several peaks, the visual system can interpret those peaks as distinct categories. Rapid categorical parsing is computationally demanding, given the severe bottlenecks of attention and working memory that allow processing of only a few objects at a time. This study tested whether such rapid parsing is automatic or requires attention, using the vMMN ERP component as an index of automatic discrimination. Twenty volunteers (16 female; mean age 22.7) took part. While participants performed a demanding central task to load attention, the researchers observed substantial vMMN responses to unattended changes in background categories defined by particular length‑orientation conjunctions. Crucially, these responses emerged under conditions where feature distributions showed multiple peaks and thus supported categorical separation. The findings indicate that spatially intermixed objects are automatically parsed into distinct categories, offering insight into how the visual system forms rich perceptual experience despite limited processing resources.