Bartolomeo Montrucchio Politecnico di Torino Dipartimento di Automatica e Informatica (DAUIN) Torino - Italy
bartolomeo.montrucchio@polito.it
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Spatial domain Spatial domain g(x) = T[f(x,y)] the neighborhood can also be 1x1
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Thresholding
Image Enhancement - 2
darkening below k and brightening above k
thresholding
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Transformation functions s=T(r) menu curves in image processing software see example with Moon image see p.675 [Zakia] for the quadrant diagram in photography
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Log transformation
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
s = c log(1+r) where c is a constant and r >= 0 a narrow range of low-level values in the input image is mapped in a wide range of output levels at right the spectrum values are now more similar and easier to be represented
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Gamma correction A CRT monitor has a gamma value of 1.8-2.5
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Gamma correction CRT
Here gamma correction means (for the CRT) to apply s=r^(1/2.5)
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Gamma correction
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Gamma correction
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Contrast stretching
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Specific range highlighting
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Specific range highlighting
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Bit-plane slicing
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Bit-plane slicing
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Bit-plane slicing
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histograms The histogram is a discrete function h(rk)=nk where rk is the kth gray level and nk is the number of pixels having gray level rk Usually is normalized p(rk)=nk/n where n is the total number of pixels Over/under-exposure are easily seen in the histogram in the digital camera display
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram equalization (I)
Spreading out the frequencies (grays levels should have a uniform density) in an image (or equalising the image) can be done for improving the image The formula for histogram equalisation is given where rk: input intensity sk: processed intensity k: the intensity range (e.g 0.0 1.0) nj: the frequency of intensity j n: the sum of all frequencies
Image Enhancement - 2
sk = T (rk ) = ! pr ( r j ) j =1 k
=! 19922008 R. C. Gonzalez & R. E. Woods
nj n
j =1
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram equalization (II)
given a probability density function (PDF) it is possible to produce a resulting uniform PDF
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram equalization (III)
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram equalization (IV)
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
10
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram equalization (V)
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram specification
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
11
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram specification 1. Mapping of the levels in the original image into sk, starting from the histogram of the original image sk= T(rk) 2. Computes G function vk=G(zk ) starting from the desired histogram as a T(zk) like sk (please note that after both applying T to the original image or G to the desired and unknown image, results should be the same) 3. Computes G-1 4. Computes the desidered levels of the image applying G-1 zk=G-1[T(rk)] zk=G-1(sk) 19922008 R. C. Gonzalez & R. E. Woods
Image Enhancement - 2
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram specification
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
12
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram specification
Image Enhancement - 2
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Histogram specification
Image Enhancement - 2
The function in (b),(2) is the inverse of that in (b),(1) It can be obtained (geometrically) reflecting the graph across the line y = x (not always it is possible) 19922008 R. C. Gonzalez & R. E. Woods
13
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Global and local histogram equalization
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Global and local histogram equalization
19922008 R. C. Gonzalez & R. E. Woods
14
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Mathematical operations Image subtraction used in segmentation and enhancement Image addition Image division multiplication of one image by the reciprocal of the other (e.g. for uneven illumination) Image averaging used for contrasting noise (e.g. in astronomy) 19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Linear spatial filtering
Given a linear, position invariant operator h(x,y): 1. Point Spread Function (PSF) h(x,y) is the response of the system to the impulse (a point of light) 2. Fourier transform of h(x,y) is H(u,v) and is the optical transfer function h(x,y) is also referred as a spatial convolution mask (only if symmetric) 19922008 R. C. Gonzalez & R. E. Woods
15
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Linear spatial filtering
Correlation +%
Convolution +%
f (x) o g(x) =
& $%
f " (# )g(x + # )d#
f (x) " g(x) =
& $%
f (# )g(x $ # )d#
f* is the complex coniugate of f (imaginery part has opposite sign)
f (x, y) o g(x, y) " F(u,v)G(u,v) f # (x, y)g(x, y) " F(u,v) o G(u,v)
Convolution theorem f (x) " g(x) # F(u)G(u) f (x)g(x) # F(u) " G(u)
Correlation Theorem (in 2D)
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Linear spatial filtering
19922008 R. C. Gonzalez & R. E. Woods
16
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Linear spatial filtering
g(x, y) =
# # w(s,t) f (x + s, y + t) s="a t="b
Example of convolution mask, where the image f has size MxN, the filter mask has size mxn, a=(m-1)/2, b=(n-1)/2 Linear spatial filtering is often referred as convolving a mask with ! an image
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Linear spatial filtering
Please note that in photography, the resulting image at the sensor level is the convolution of the original (perfect) image with the finite aperture of the lens (diaphragm) if the original image is a very little point (e.g. 1 m) the result is as in figure if the lens is out of focus the result is much larger (so called bokeh) In figure there is the PSF of an optical system (a telescope with f=2500 mm and d=120mm), given as input a point of light (an artificial star of 35m at more than 15m)
112.5 m (core is 22.5 m) for a ~f20 lens, seen through a 3.75m pixel-wide CCD 19922008 R. C. Gonzalez & R. E. Woods
17
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Averaring filter masks
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Averaring filter masks
19922008 R. C. Gonzalez & R. E. Woods
18
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Averaring filter masks
19922008 R. C. Gonzalez & R. E. Woods
Digital Image Processing, 3rd ed.
Gonzalez & Woods www.ImageProcessingPlace.com
Image Enhancement - 2
Non linear filtering Median filter
median is the numerical value separating the higher half from the lower half median lter computes the median under the mask (here 3x3) and uses it as a resulting value it is very effective for some images 19922008 R. C. Gonzalez & R. E. Woods
