Escolar Documentos
Profissional Documentos
Cultura Documentos
References
1. Gonzalez and Woods, “ Digital Image
Processing,” 2nd Edition, Prentice Hall,
2002.
2. Jahne, “Digital Image Processing,” 5th
Edition, Springer 2002.
3. Jain, “Fundamentals of Digital Image
Processing,” Prentice Hall 1989
03/25/11 2
Overview
• Human perception (focus of this
discussion)
– Machine perception (ocr)
• Application specific
• Heuristic based: result better than the
original image – subjective assessment
• Spatial vs frequency domain
03/25/11 3
Spatial Domain
• Based on the collection of pixels in the
image
• Enhancement techniques will yield
– Noise reduction
– Neighborhood smoothing
– Highlighting of desired feature
03/25/11 4
Spatial Domain – Math Framework
From [1]
From [1]
03/25/11 6
Larger Neighborhoods
• Objective – determine g(x,y) based on
input intensity (gray level) values f(x,y) in
the neighborhood of x,y.
– Mask processing or filtering
– Each of the elements in the neighborhood has
an associated weight
– g(x,y) depends on f(a,b)|a,b є N(x,y)
03/25/11 7
Basic Gray Level Transformations
Light
From [1]
Dark
Dark Light
03/25/11 8
Image Negatives
From [1]
03/25/11 9
Log transformations
• s = c log (1 + r)
• r >= 0; hence 1 + r > 0; log 0 = ?
• Log transformations are useful, when there is a large dynamic
range for the input variable ( r ).
From [1]
Stretch lower
(ldarker) gray levels
03/25/11 11
Display distortion correction
1. Gamma correction
can also fix the
distortions in color.
2. More important with
the internet.
3. Many viewers, variety
of monitors.
4. Gamma of view
station is not known.
5. Preprocess using an
average gamma.
From [1]
03/25/11 13
Piece-wise Linear Transformation
Contrast Stretching
From [1]
From [1]
03/25/11 15
Bit plane Slicing
MSb (bit 7)
From [1]
03/25/11 17
Histograms for 4 images
From [1]
s
Applying transformations to histograms, can use results from
probability. s1
Consider the transformation S = T( r ); 0 <= r <=1 which has the
following properties: r
(a) T( r ) is single-valued and monotonically increasing in [0,1]
(b) 0 <= T( r ) <= 1 for r in [0,1] r
Single valued condition ensures that an inverse transformation
exists, and the monotonic condition ensures the increasing
order of the pixels from black to white.
(a) and (b) do not ensure that the inverse transform is single
valued. s1 s
03/25/11 19
Histogram Equalization
From [1]
03/25/11 21
Mapping for Histogram
Specification
03/25/11
From [1] 22
Example of Histogram Specification
From [1]
03/25/11 23
Continued
From [1]
03/25/11 24
Localized Histogram Equalization
From [1]
03/25/11 25
From [1]
03/25/11 26
Histogram Stats
From [1]
03/25/11 27