img/" width="727" height="1090" useMap="#Map">
CognitivePsychology ­ PSY 504
VU
Lesson 14
PATTERNRECOGNITION
We know how information is first recorded by senses andselected for processing by attentional
mechanisms.The most criical questionfor a theory of perception is how is thisinformation
recognizedfor what it is?
Mostresearch has focused on letter recognition or recognition of letters identity: forexample how
do we recognize
A
as an instance of
A
TemplateMatching Models
Thetemplate-matching theory of perceptionassumes that a retinal image of an object is faithfully
transmitted to the brain. An attempt is made to compare it directly to various storedpatterns.
Thesepatterns are calledtemplates. So, the basic is that the perceptual systemtries to compare
theletter to templates it hasfor each letter andreports the template thatgives the bestmatch.
Figureillustrates various attempts to make template matching work. In each case, an attempt is
made to achieve a correspondence betweenthe retinal cells stimulated by the A and theretinal
cellsspecified for a templatepattern.
Inputrepresents retinalcells.
Templatematching
Thepattern is stored for eachretinal cell activated; if different retinal cells areactivated there
won't be a match. As the figure is shown:
40
img/" width="727" height="1090" useMap="#Map">
CognitivePsychology ­ PSY 504
VU
a
successfulmatching
b
wrongmatching
c
successfulmatching
d
wrongpart of retina
e
wrongsize
f
wrongorientation
g
and h non-standard images
Examples in machines
Sometemplate matching machinesare;
Finger-printmatching machines. Fingerprints machine are templatematching machines.These
machinesmatch finger prints withcriminal fingerprints.
Barcode reading machines in shopping stores are alsoexample of template matchingmachines.
ATM/ Credit card readingmachines. These machinesmatch the numbers of ATM card with
passwords.
Butthese are machines thatrecognize exact matches. Any minor change means there is no
match.
Humanflexibility
In Humanswe canrecognize LARGE letters andsmall characters, Characters in the wrong place
and in strange.
It means if you meet youruncle after many yearseven he is in different appearanceyou can
recognizehim. This is humanflexibility to recognize things,people or places.
Problemswith templates
Humanperceptual system is muchmore flexible.
Mostpatterns we encounter undergo at least some minor changes.Like,
Lettersare different font sizesand styles
Shapeschange slightly
Orientationsare different
Butstill we can recognizethings.
FeatureAnalysis model
In this model, Stimuli arethought of as combination of elemental or primitive features.
Thefeatures for alphabets mayconsist of horizontallines
_
verticallines
I
41
img/" width="727" height="1090" useMap="#Map">
CognitivePsychology ­ PSY 504
VU
Lines at approx 45 degree angle/
And curves (
Likealphabet  T  hastwo line one verticaland other is horizontal lineboth make alphabet
T. Allletters are consisted on these four patterns. Ourmind analyzes lettersaccording to
thesefeatures.
How is it better than templatemodel?
Featuresare mini-templates butfeatures are simpler.
Onecan identify criticalfeatures for a letter
For A, 2 45 degree angles (/ \) intersect at the top and a horizontalline _ intersects
both of these near the middle.the pattern of A consists of these lines plus a specification
as to how they should be combined.
Thesefeatures are very muchlike the output of edgeand bar detectors in thevisual cortex.
Example
Many other details areunimportant
So,
A A A A A A AAA
All of the above can be recognized as the sameletter. Even they are in different shapes and
differentstyles.
In the feature model we do notneed a template for eachletter but only forevery feature, this
would be a great saving. We have 26 letters small and capital. If letters have manyfeatures in
common,subjects are prone to confuse them.
By these four features we canstore all 26 letters.English model is alsoapplicable in Urduletters
withthe addition of Dot (nukta).Urdu writing is much morecomplex than English. Urdureading is
alsodifficult as well.
Thefeature model has a number of advantages over thetemplate models.
First,since the features aresimpler, it is easier to seehow the system mighttry to correct forthe
kinds of difficulties caused by templatesmodel.
A second advantage of thefeature combination scheme is that it is possible to specifythose
relationshipsamong features that aremost critical to thepattern.
42