Despite extensive research and experimentation in the area of computer vision, many issues and complications are hindering progress in the field. These issues begin at the most fundamental level: the defining of terms, starting with the term Computer Vision. What exactly is meant by Computer Vision and what is its purpose? The simple answer is that one purpose is object recognition. This raises other issues about the specific types of recognition. Is it enough to simply isolate an object in a picture (syntactic recognition) or should the computer be able to conceptualize and make sense of the image (semantic recognition). And what is considered an object? Perhaps people, desks, chairs... does an entire room qualify as an image that the computer should recognize? Should the computer be able to pick out each blade of grass in a field?

Another limiting factor has to do with time complexity. The average size of an image is around 500,000 pixels (half a million!). Think about the time necessary to do just edge detection on an image of that size. Then add recognition or depth recovery. It requires an enourmous amount of processing -- something today's computers are barely up to. With the technologogy of today we have trouble processing these images in real time, so real-world applications for Computer Vision are limited.

The images themselves can complicate matters. One of the largest complications in an image is noise. We discussed earlier some ways of dealing with noise, but it's not always easy to do. Lighting, shadows, illuminations, textures, and dirt all constitute noise in an image. The problem arises between desirable and unwanted noise. How can the computer know to filter out the undesirable noise without losing the important information? Other difficult pictures are impressionist paintings by artists such as Monet and Surat. Up close, the paintings look blurry -- almost meaningless -- as they are composed of many seemingly random dots that, from afar, look like a lucid picture. Could a computer ever pick out these impressionist images?

As to the methods employed to simulate computer vision, some say we should model computer vision after human vision. The problem is, though, that human vision itself is inaccurate. Consider the following pictures. These are eye puzzles and illusions -- images designed to play tricks on our eyes.

Do the diagonal lines look parallel in the right-hand image? They are!

Who do you see in this image? An old woman, a young one, or both?

The diagonal line may appear to bend behind the rectangle, but it's straight.

Nonetheless, there are complications in every field of research. Indeed, if there were no complications, there would be no need for research. Given the progress that has been made and the potentional possible results which could stem from further study, we continue to search for better ways to complete this task of Computer Vision.

Back to the Table of Contents.