Browse Prior Art Database

Computer Simulation of Visual Error Detection in Images

IP.com Disclosure Number: IPCOM000106194D
Original Publication Date: 1993-Oct-01
Included in the Prior Art Database: 2005-Mar-20
Document File: 4 page(s) / 239K

Publishing Venue

IBM

Related People

Lightstone, SS: AUTHOR

Abstract

Described is a software design called Computer Simulation of Visual Error Detection in Images (CS-VED) that performs high-performance fuzzy comparisons between raster images, to determine similarity between images, using logic that simulates human vision and perception.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 25% of the total text.

Computer Simulation of Visual Error Detection in Images

      Described is a software design called Computer Simulation of
Visual Error Detection in Images (CS-VED) that performs
high-performance fuzzy comparisons between raster images, to
determine similarity between images, using logic that simulates human
vision and perception.

      In the context of CS-VED, two images are considered similar if
and only if the images would be considered to be essentially the same
when viewed by a human observer.  For example consider two bilevel
images of white noise.  Such images are actually similar, since
almost any observer would evaluate noise being similar to noise, even
though on average only half of the picture elements match.  Had one
of those images been a page with one half completely black and the
other side white, fifty per cent of the picture elements would still
match.  Yet the images would be obviously dissimilar, since no
observer would consider them even remotely the same.  This definition
of similarity is deliberately vague, since the tolerance of human
observers is dependant on the accuracy to which they examine images.
Sometimes very little difference will be tolerated, while at other
times, the allowable difference for similarity will be large.  The
range is usually dependant on the requirements/needs of the
application or individual.

The following assumptions are made within the algorithm:

      Image size - The two images must be exactly the same size, and
have the same approximate orientation.

      Positional probability of data element detection - The
probability of finding a mismatch data-element in the source data
highest is near the location of the current data-element.  This
assumption can be understood in a very practical way.  When we look
at images, we perceive them as similar by the positional similarity
of the colours, edges, and artifacts.  The more images are similar,
the more these factors will match.  For images that generally do
match, the position of these characteristics (colour, edges, etc.)
should be similar, and where one finds an edge in one image, it is
most probable that the same will be detected near the same location
on the other image.  The further one must stray in searching from
that site to find the edge, the less likely it is that the edge will
be found.

CS-VED uses the following simple algorithm:

1.  Start with two images i) Reference, which represents the control
    image, and ii) Secondary, which is the image that is being
    compared to the Reference.

2.  Define a percentage of tolerable noise that may be allowed when
    determining similarity.  Typical values may be 0.05% for bilevel
    raster data.  This percentage multiplied by the number of
    discrete data points is called NOISELIMIT.  It represents the
    total amount of noise that is allowable for an image to be
    considered SIMILAR to the Reference.

3.  Define a maximum per...