Computer Vision Applications
Future
Despite the enormous difficulties inherent in making a computer "see," the things
which could be done with a sighted computer are too cool to let us stop trying.
In the future (as suggested by Lawrence Stevens in his book Artificial
Intelligence pg. 47):
- we could place robots in factories to inspect items, detect defective material, and discard that defective material.
- we could rely on computers for medical purposes -- they could review x-rays and blood samples to look for irregularities, potentially reducing the risk for mis- (or missed) diagnosis.
- a computer / camera system set up in a plane could view remote areas from above and then use those images to create maps of them.
- sighted missiles could be used in military tactics rather than heat-seeking ones.
- we could set up traffic control cameras above highway systems or city streets to provide information about where cars should go for the least congested route.
- computers could aid blind people in everyday tasks.
- Oil and other such resources can sometimes be detected by examining the surface of a location. Computers could be used to detect these resources in remote areas.
We could have robots in our houses sort our laundry into colors and whites. Security cameras could be made to alert store owners if a theft was taking place. The list goes on and on. It's too tempting to think that there's a possibility of these ideas becoming reality not to do research in computer vision, no matter how hard the task may seem!
Present
Fortunately, we have made sufficient progress in the field of computer vision for
applications which are actually in use today. As mentioned by Ching Y. Suen in
the Proceedings of the Third International Conference on Document Analysis and
Recognition, the techniques of computer vision are used in:
- bar code scanning in supermarkets
- processing utility bills
- sorting mail
- processing bank checks
- reading financial data and business forms
- extracting symbols and information from maps and engineering drawing and other areas as well.
Many of these applications rely on text recognition and often handwritten data
recognition. Much research has been done in the creation of new techniques for
computer recognition of handwritten data in particular.
- Cheque Reading
One paper (included in Proceedings) by Jean-Pierre Dodel and Rajjan
Shinghai of Concordia University, entitled Symbolic / Neural Recognition of
Cursive Amounts on Bank Cheques, describes (in limited detail) their method
for computer recognition of the alphabetic amount written on a cheque.
Instead
of relying on letter-recognition, they instead begin with a vocabulary of 31
words (including one, ten, twenty, hundred, dollars, etc.) and rely on
word-recognition. They have created a Symbolic Classifier which looks for loops,
ascending and descending scribbles (such as the top of a 't' and the bottom of a
'g'), capital letters, and other particulars. A Neural Classifier then is used
to recognize words segments, i.e. for the word 'Two,' the Symbolic Classifier would find 'T' and
the Neural Classifier the 'wo.' Words then are either entirely recognized by the Symbolic Classifier, such as
eight, whose descending 'g' is enough to mark the word as an 'eight;' recognized
partly by the Neural Classifier, such as Six, whose 'S' is recognized by the Symbolic Classifier and whose 'ix'
is recognized by the Neural Classifier; or recognized entirely by the Neural Classifier, such as 'nine.' In
testing, the computer was trained by 80 people's handwritten data, and upon
testing by another 40 volunteers, was able to recognize 70% of the data.
There are other ways for handwriting recognition -- indeed, a check is not
generally a complicated image whereas text on a diagram may be more difficult for
a computer to understand. This paper is an example that most work in computer
vision is application-specific.
Back to the Table of Contents.