A webinar on " Role of low-level visual features in behaviorally-relevant face categorization"
Dr. Bhuvanesh Awasthi is a cognitive scientist. His research focused on mechanistic and functional accounts of perception, emotion and decision-making in humans. His early training was from the University of Pune, a Consciousness Studies Masters from BITS Pilani and a PhD from Sydney, Australia. He has been invited to present his research to a variety of audience across North America, Europe, Middle East, Russia, China, India & Australia.
We recently got an opportunity to attend a webinar by Dr. Awasthi titled as "Investigating the role of low-level visual features in behaviorally-relevant face categorization" which was organized by a prestigious HEI of Jodhpur.
The highlights of the webinar were as follows:
Mr. Awasthi has majorly worked in Neuroscience. The crucial issue in Cognitive neuroscience which has been discussed here is the extraction and processing of relevant information from the visual environment.
Cognitive Neuroscience provides us the standard framework of Visual Perception.
Low Level Information is assumed to be an impediment need.
Taking an example of a picture of a Panda, he explained that Low Level Information is processed at the retinal level. This information is not fully carried forward. Only some aspect is forwarded.
It is not carried forward because of Spatial Frequency. The Spatial Frequency is the variation in luminescence. Low level features are at the early encoding stage. The Low Spatial Frequency gives the coarse details and the High spatial Frequency gives the finer details. Then we combine both these details in our visual system to form the image.
The faces are processed at a faster level. The gender, age, expression, intention, all these are found by looking at the face of a person. Within a rapid glance, we get a lot of information.
The faces are perceived holistically. In holistic approach, the processed part is taken in a single glance rather than a collection of independent features. It's not taken in parts.
Holistic and Analytical are two types of facial processing. In analytical, part based face processing is done.
Spatial Frequency -
The spatial frequency is of two kinds, Low spatial frequency ( LSF ) and High spatial frequency ( HSF ).
The LSF is channelized through magnocellular pathway. The HSF is received by parvocellular pathway.
Dr. Awasthi further talks about the Magnocellular advantage model of visual perception where magnocellular has a coarse resolution and parvocellular has a finer resolution, He says Neuroscience needs behavior as converting a reductionist bias.
Cognitive neuroscience research is about understanding brain and behavior. The aim is to decipher causal explanation through neural path,
The focus is on Spatial Frequency Information.
Now the question is that "what happens when both ( LSF AND HSF ) spatial frequency information are available?"
Then we get a Hybrid Image.
This he explained through an example, where we were asked to look at the desktop screen and find out what image wee see on the screen.
It was observed that when we moved close to the screen, it appeared to be an image of a male.
As we moved away from the screen, it appeared to be an image of a female.
So, closer to the screen the dominant perspective was the male face and away from the screen the dominant perspective was the female face. It depends upon how many fine or coarse details we have.
Based on the LSF and the HSF, a hybrid image was created with the combination of male and female face images.
There were four types of images:
M + M
F + F
These were both, the Congruent condition.
These were both the In congruent condition. Here, both the images were superimposed.
Further Dr. Awasthi mentioned about a recent study using transcranial Random Noise Stimulation in which AC Current was used.
In Dr. Awasthi's study, two hybrid faces were presented in alternate blocks at fovea and periphery.
Stimulation - EEG 10-20 system with Center point of active electrodes placed over F8 location.
Stimulus- Induced via a BrainSTIM stimulator.
For active high frequency ( 100-640 Hz spectrum ). tRNS delivered online, 1mA current was used.
The question was to find out which image is male or female.
So, firstly the two images were taken and kept adjacent to each other. That was the fovea condition,
Then, two images were taken and kept far away from each other at a distance. That was the peripheral condition,
The response time was taken due to tRNS. The fovea had a response time significantly faster than periphery. Congruity and Conflict was significantly faster.
They found a two - way interaction between tRNS (Sham and Stimulation) and TargCon.
When target was at Incongruent condition, then response time was longer. Thus, Stimulation affects Target congruity as well.
There was a three - way interaction between tRNS (Sham vs Stimulation) x Eccentricity (Fovea vs Periphery) x TargCon --- approaches significance.
It was expected that Interference by LSF, would be larger at Periphery than at the Fovea.
tRNS increases cortical excitability.
Sources of Error-
Both magnocellular and parvocellular might receive input (not experimentally proved).
Hybrid Study Conclusions-
Spatial Frequency - low level stimulus.
LSF information in faces is behavioral.
Results demonstrated a behavioral significance and the neural primacy for low spatial frequency information in the perception of faces.
After Dr. Awasthi's demonstration, several questions were put forward by the Moderator on behalf of the audience.
One of the questions addressed by Mr. Awasthi was "since, the simulation of this technique leads to reduce response time, so can this technique be used in training?"
He responded that this technique can be used where Face Perception is compromised.
Next question put forward to him was "Can there be any application of this technique?"
For which he answered that they still have to do experiment and check it but in his opinion it can find an application in Passport / ID Control mechanisms.
Mr. Awasthi shared his view that this technique can be used at the airports where Real Face Matching is done with the picture. He says, in Australia and UK, people who are checking the passports, they don't recognize the person. May be because, sometimes the passports are 10 years old or other likely reasons. Nowadays, more and more work is being done by Machine. So, in Human Face Perception, we can use this technique.
Next question addressed was that "Post Covid, does it add to more Research Potential?"
For this he responded that, it is the Human Capability to see the face of a person and find out the Gender, Age and Expressions of the other person. Now, Post Covid, people would be having masked faces. So, expression identification would be difficult when half of the face can not be seen. Face expressions which we detect are from Eyes and the changes in mouth. They both combine and tell us the expressions of a person. For certain expressions, we can see the disgust / threat in a person's eyes. Due to masked faces, this will not be clear Post Covid. So, owing to the societal changes, it definitely adds to more Research Potential.