Computer Vision — Not Only Luck

Today I continue the short posts on exciting application areas related to things that we do at ArrayFire. Computer vision is an area that continues to explode in useful applications.

The phrase “computer vision” was formulated to describe the general goal: to enable computers with attached cameras to intelligently “see” the same things in those images that humans do. For instance, humans are able to visually distinguish humans from cars in traffic scenes or distinguish cats from dogs in animal scenes.

Computers need algorithms to be able to “see” like humans. Algorithms enable the computer to process the pixel values (which are numbers to a computer) and identify the salient features. It is an enormous challenge and people spend their whole careers trying to teach computers to find faces in photos automatically or to find tumors in MRI scans automatically.

Much progress has been made in computer vision recently. Competent algorithms exist for many types of problems. These algorithms can now perform reasonably fast given advances in heterogeneous computing.

Computer vision is still not fully developed. Computers are still unable to see as well as humans in most cases. Computer vision is most powerful when it is coupled with machine learning to find objects in incoming imagery (computer vision) as well as match those objects against a large body information (machine learning).

What are your thoughts on computer vision?

Related articles