Assigment 8
in GEG2210
- Data Collection - Land Surveying, Remote Sensing and Digital
Photogrammetry
Image enhancement, filtering and sharpening
By Petter Reinholdtsen and Shanette Dallyn, 2005-05-01.
Logged into jern.uio.no using ssh to run ERDAS Imagine. Started by
using 'imagine' on the command line. The images were loaded from
/mn/geofag/gggruppe-data/geomatikk/
Tried to use svalbard/tm87.img, but it only have 5 bands. Next tried
jotunheimen/tm.img, which had 7 bands.
The pixel values in a given band is only a using a given range of
values. This is because sensor data in a single image rarely extend
over the entire range of possible values.
Evaluation of the different bands
band 1, blue (0.45-0.52 um)
Visible light, and will display a broad range of values both over
land and water. Reflected from ice, as those are visible white and
reflect all visible light waves. Histogram show most values between
30 and 136. Mean values of 66.0668. There are one wide peak with
center around 50. There are two peaks at 0 and 255.
band 2, green (0.52-0.60 um)
Visible light, and will display a broad range of values both over
land and water. Reflected from ice, as those are visible white and
reflect all visible light waves. Histogram show most values from 8
to 120. The mean value is 30.9774. There are two main peaks at 20
and 27. There is also a pie at 0.
band 3, red (0.60-0.69 um)
Visible light, and will display a broad range of values both over
land and water. Reflected from ice, as those are visible white and
reflect all visible light waves. Histogram show most values from 33
t 135, with one wide peak around 52. There are also seem to be two
peaks at 0 and 255. The mean value is 34.3403.
band 4, near-infraread (0.76-0.90 um)
Water acts as an absorbing body so in the near infrared spectrum,
water features will appear dark or black meaning that all near
infrared bands are absorbed. On the other hand, land features
including ice, act as reflector bodies in this band. The histogram
show most values between 7 and 110. The mean is 40.1144. There are
two peaks at 7 and 40.
band 5, mid-infrared (1.55-1.75 um)
The ice, glaciers and water do not reflect any mid-infrared light.
The histogram show most values between 1 and 178. The mean is
49.8098 and there are two peaks at 6 and 78, in addition to two
peaks at 0 and 255.
band 6, thermal infrared (10.4-12.5 um)
Display the temperature on earth. We can for example see that the
ice is colder than the surrounding areas. The histogram show most
values between 36 to 122. The mean is 102.734. There are one wide
peak around 53, in addition to two peaks at 0 and 255.
band 7, mid-infrared (2.08-2.35 um)
The ice, glaciers and water do not reflect any mid-infrared
frequencies. The histogram show most values between 77 and 150.
The mean is 24.04, and there are one wide peak at 130 and a smaller
peak at 83, in addition to one peak at 0.
Image enhancement
-----------------
We can get a good contrast stretch by using the histogram
equalisation. This will give us the widest range of visible
separation between features.
Displaying colour images
------------------------
Comparing a map we found on the web,
and the standard infrared image composition, we can identify some
features from the colors used:
- water is black or green
- ice and glaciers are white, while snow is light green.
- vegetation is red.
- non-vegetation is brown or dull red when closer to snow and
glaciers.
Next, we tried to shift the frequencies displayed to use blue for the
red band, green for the near ir band and red for the mid ir (1.55-1.75
um). With this composition, we get some changes in the colours of
different features:
- water is black
- ice and glaciers are light blue, while snow is dark blue.
- vegetation is light green and yellow.
- non-vegetation is red or brown.
Filtering and image sharpening
We decided to work on the grey scale version of the thermal infrared.
This one has lower resolution then the rest of the bands, with 120m
spatial resolution while the others have 30m spatial resolution.
The high pass filtering seem to enhance the borders between the
pixels. Edge detection gave us the positions of glaciers and water.
We tried a gradient filter using this 3x3 matrix: [ 1 2 -1 / 2 0 -2 /
1 -2 -1 ]. It gave a similar result to the edge detection.
We also tried unsharp filtering using this 3x3 matrix: [ -1 -1 -1 / -1
8 -1 / -1 -1 -1 ]. This gave similar results to the edge detection
too.
We started to suspect that the reason the 3x3 filters gave almost the
same result was that the fact that the spatial resolution of the
thermal band is actually 4x4 pixels. Because of this, we tried with a
5x5 matrix, making sure it sums up to 0.
-1 -1 -1 -1 -1
-1 -1 -1 -1 -1
-1 -1 24 -1 -1
-1 -1 -1 -1 -1
-1 -1 -1 -1 -1
Next, we tried some different weight:
-1 -1 -1 -1 -1
-1 -2 -2 -2 -1
-1 -2 32 -2 -1
-1 -2 -2 -2 -1
-1 -1 -1 -1 -1
This one gave more lines showing the borders between the thermal
pixels.
From: shanette Dallyn
Subject: Re: My notes from todays exercise
To: Petter Reinholdtsen
Date: Sat, 30 Apr 2005 15:16:59 -0400 (EDT)
Hey Petter!
Allright, I looked up some stuff on statistics and the most valuable
conclusion that I can come up with for the histograpm peak question is:
"The peak values of the histograms represent the the spectral sensitivity
values that occure the most often with in the image band being analysed"
For the grey level question go to http://www.cs.uu.nl/wais/html/na-dir/sci/
Satellite-Imagery-FAQ/part3.html I found this and thought that the first major
paragraph pretty much answered the question for the grey levels.
theory of convolution:
Specialty Definition: Convolution
(From Wikipedia, the free Encyclopedia)
In mathematics and in particular, functional analysis, the convolution
(German: Faltung) is a mathematical operator which takes two functions and and
produces a third function that in a sense represents the amount of overlap
between and a reversed and translated version of .
The convolution of and is written . It is defined as the integral of the
product of the two functions after one is reversed and shifted.
The integration range depends on the domain on which the functions are
defined. In case of a finite integration range, and are often considered as
cyclically extended so that the term does not imply a range violation. Of
course, extension with zeros is also possible.
If and are two independent random variables with probability densities and ,
respectively, then the probability density of the sum is given by the
convolution .
For discrete functions, one can use a discrete version of the convolution. It
is then given by
When multiplying two polynomials, the coefficients of the product are given by
the convolution of the original coefficient sequences, in this sense (using
extension with zeros as mentioned above).
Generalizing the above cases, the convolution can be defined for any two
square-integrable functions defined on a locally compact topological group. A
different generalization is the convolution of distributions.
I hope this will help!
Shanette
Petter Reinholdtsen
Last modified: Sun May 1 14:28:48 CEST 2005