Programming for Artificial Vision

Programming for Artificial Vision

I have been curious about technological advancements in artificial vision, so I did a little bit of research to see how computers are seeing these days. As it turns out, researchers have been able to instill new programming into computers that allow the systems to recognize optical illusions that are context dependent.

Optical Illusions

I find it fun and interesting to look at optical illusions and to learn about why our brains see things the way we see them. At the same time, this is a level of understanding that is more than just academic. It takes a great deal of research to understand.

The class of optical illusions known as contextual phenomena are all dependent on context. As an example, you might see a circle as a certain color but if there is another colored ring around it, you will see it as a different color or different shade.

Computers Seeing Optical Illusions

At Brown University, a group of computer vision technologists decided to learn about the neural mechanisms of context-dependent phenomena. I learned that they studied this based on the idea that optical illusions are not a glitch of the mind but rather a feature that computers could learn.

The point of the study was to learn more about how the brain sees so that these algorithms could be programmed into computer vision systems. It used a computational model that is constrained by the neurological limitations of the visual cortex, thus allowing for optical illusions to be recognized.

An Innovation Included

I found out that the research team included a certain pattern of feedback connections which are hypothesized to be between neurons. I learned that these connections for feedback are capable of increasing or decreasing neuronal responses and this is dependent on visual context.

Since these feedback connections are not at all included in learning algorithms for deep learning, there is a limitation there. Deep learning algorithms usually just include feed forward connections and not feedback connections. Now, with this inclusion, the systems could have better visual distinction abilities.

Testing it All

It was interesting for me to discover that, once the model was made up, the research team did indeed present the computer with context-dependent illusions. Researchers then tuned the feedback strength accordingly and this resulted in the computer being able to distinguish between optical illusions present.

After testing on even more optical illusions, they were able to create an algorithm that can see illusions the way that humans do. To me, this means that we are getting much closer to having artificial vision systems that can see much in the way that humans can and this may be able to help those with serious visual disabilities.

The goal here is to improve artificial vision technology. It seems to me that scientists are now on the brink of not only understanding exactly how the brain sees but also how to create computer programs and technology that will see in the same way.

It is a matter of very precise programming and we are gaining the skills to do this.