A lot of coffee has been spilled over artificial intelligence for the last five decades. The mission of creating machines with independent cognitive abilities has kept the best minds of the human world occupied for years. Computer vision that is nothing but the ability of computers to extract information out of images was one of the less-discussed sub-fields of the AI family. A number of scientists thought computer vision would be as simple as attaching a camera to a computer. For a machine that can learn the entire gamut of musical patterns in Beethoven’s work in a few hours, how difficult can it be to distinguish an elephant from a floor mat?
However, as it turns out, forty years into incessant research and scientists cannot still make the computer really see.
Complexity of vision
We never had to think about vision it was just given to us. We see an object once and the next time we recognize it irrespective of the time, place, lighting, temperature or any other variables. It comes so naturally to us that we never really considered the complexity of vision. Light reflects on things, an inverted image is captured in the eyes, the optic nerves transmit the image to the brain, it mirrors the image, interprets it, makes us understand and devises a response; neatly done. Although we know this much, we have never really been able to organize the pieces well enough to replicate the process to full effect.
Computer vision is different from image processing
Image processing is another subfield of AI that can be included in an applied AI machine learning course. But image processing is quite different from computer vision. The former deals with restoring or recreating images; changing photometric features like brightness or color; replicating or enhancing a previously existing image, etc. It does not involve understanding the content of the images or the digitally portrayed objects which is what computer vision is all about. While image processing has come a pretty long way as you would already know from all the fancy photo augmenting applications on social sites, computer vision has yet to reach its full potential.
The current status
Of late there has been rapid development in the field of computer vision. Thanks to the effective application of neural networks, computer vision can now be used for a number of tasks. Machines can now be taught to recognize certain objects and specific characteristics of those objects. They can also perform tasks such as ensuring automotive safety, merging graphic imagery with the movement of actors in films, surveillance, and fingerprint recognition. As a subfield of AI computer vision is burgeoning, to say the least.
