The human visual system is a sophisticated machinery that allows us to process color images and also recognize objects in black and white images. Now a study of Massachusetts Institute of Technology (MIT) offers a possible explanation for how the brain develops this ability.
The research also explains why identifying objects in black and white images is more difficult for people who are born blind and then regain sight after surgery.
The results of the study, whose conclusions were published this Thursday in the journal Science, highlight the importance of the different stages of the development of the vision human and will help develop ways to improve computer vision systems.
Human vision is formed in several stages. It is believed that the vision of newborns is very blurry and colorless and that as they mature, their acuity improves. After two months, the you drink They begin to perceive colors, particularly red-green contrasts.
But today, it is still not known whether acquiring the human ability to distinguish colors progressively has advantages.
Learning from babies or learning from children
To find out, an MIT team led by Marin Vogelsang and Lucas Vogelsang used data from Project Prakash, a scientific initiative created in 2005 by MIT Professor of Brain and Cognitive Sciences Pawan Sinha, which surgically restores the sight of adolescents. of the India with dense bilateral cataracts.
The team gave the children a simple object recognition test with color and black and white images and, compared to a group of children with normal vision, found that they had more difficulty recognizing images of common objects on a scale of gray than in color.
The team hypothesized that the late color development experienced by children with normal vision helps them achieve robust recognition of objects in the face of color variations, while children who do not have this stage of visual development and who They learn to see colors directly – like the children operated on in the Prakash Project – they rely more on color perception when identifying objects.
To investigate this idea further, they used simulations with deep neural networks, one trained to recognize grayscale and color objects, and another that only understood color images.
They found that the model inspired by human visual development could accurately recognize objects in any type of image and also with color manipulations, and that the other model, more similar to the experience of the children of the Prakash project and trained only with images in color, was not good with grayscale images or manipulated tones.
Brain plasticity
The study explains that models that start with grayscale input learn to rely on light to identify objects and that when they start receiving color information, they do not change their approach much because they have already learned a strategy that works well, while Models that start with color images and then add gray scale no longer learn to be as accurate as the first ones.
Researchers believe that something similar occurs in the brain human, which is more plastic in the early stages of life and easily learns to identify objects based on luminosity alone.
When newborns receive very limited information about color, the brain is forced to learn to distinguish objects based on the intensity of the light they emit, rather than their color. Later, when the retina and cortex are more developed to process colors, the brain incorporates chromatic information, although it maintains its previous ability to recognize images without depending only on color.
The team concludes that the scarcity of information about color may be beneficial for the developing brain because it learns to identify objects from scarce information, the authors believe.