Ben Schneider

Zach Sorenson

Neil Agarwal

Bailey Zhang

Vision Recognition Handout

Vision Recognition is the field of computer science that deals with machine vision; interpretation of real world images by computers to be used for a variety of reasons.  Some reasons include: help for people with disabilities, virtual reality, detection of pests, more practical user interfaces, etc.  Though vision recognition is not a new field, it has yet to reach its full potential.

Optical Character Recognition (OCR) has been around for several years, but it is still not as completely effective.  OCR is a way to convert text from a printed source (such as a sheet of paper with printer text or hand written text) to text that can be used in a word processor so the user can manipulate the text.  This would be useful if you had a printed copy of a document that you wanted to edit, but you didn't have a copy of the file to open.  There are three types of OCR methods: character-based word recognition (a three-step method), segmentation-based word recognition (a two-step method), word-shape recognition (a three-step method).  Character based recognition is the most effective at recognizing characters correctly, with word-shape recognition being the least effective.  Calera Wordscan Plus 1.0, Caere Omnipage Professional 3.0, Xerox Imaging Systems AccuText 3.0 are examples of OCR software available to the public right now.

Augmented reality is still in an early stage of research and development at various universities and high-tech companies.  Eventually, possibly by the end of this decade, we will see the first mass-marketed augmented-reality system, which one researcher calls "the Walkman of the 21st century."  What augmented reality attempts to do is not only superimpose graphics over a real environment in real-time, but also change those graphics to accommodate a user's head- and eye- movements, so that the graphics always fit the perspective.  Here are the three components needed to make an augmented-reality system work: head-mounted display, tracking system, mobile computing power.

Vision recognition has been successfully implemented in many fields.  In industry, it is used in combination with robotics for product inspection.  In the movies, it is used for motion capture and the animation of characters.  In personal computer use, it is implemented in OCR.  Overall, vision recognition is used in agriculture, architecture, character recognition, cultural fields, forensic science, medical science, image processing, reverse engineering, military technology, navigation, remote sensing, safety, and sports.

There are many current projects in vision recognition.  It is a very important field, to robotics, AI, and generally, all future PC's.  These projects are varied, but a common project is content recognition.  The computer turns a digital image into component shapes, querying a database them to discover their purpose.  Another project involves reading blurred text from a television image, and clarifying them automatically.  Other projects involve human nude recognition, and hand gesture recognition.  In addition, there are many, many other projects currently underway in the field of vision recognition, computer vision, and machine vision.

Generally, vision recognition is used for creating useful autonomous agents that will be able to interact in the real world, making it easier to communicate with humans via things like gesture recognition and handwriting recognition, passively observing and analyzing images more efficiently than humans, and better understanding biological vision.  Probably the biggest problem that vision researchers currently face is an incomplete understanding of our own vision.  Currently, most vision recognition techniques involve mathematical algorithms.  Two major problems researchers face are converting pixels into shapes, and making a 3-dimensional representation of a 2-dimensional image.  Mathematical edge and shape recognition algorithms are generally used to get around the first problem.  For 3D representation, researchers have been trying out binocular vision systems, although those are still in their experimental stages.