Assignment 2 : Filtering, Enhancement, and Extraction
Professor: Tarmo Remmel
TA: Yikalo Araya
TA: Yikalo Araya
Student Name : Jinha Bok
Student #: 209216631
Data: Feb 4, 2011
Monday, January 24, 2011
Question 1 Image Detail and noise detection
[Part 1] Explaination of Noise in Each Band
[Noise]
The noise is particularly visible over homogeneous surfaces such as clouds and waterbodies. In addition to the Atmospheric noise, Band 1 (TM1), band 2 (TM2) and band 3 (TM3) are experencing some atmospheric noises and sensor artifacts. But band 4 and 5 do not have any noises and artifacts.
 |
TABLE 1-1 it shows which band contains what types of noise, and this has discripstions of each noise that each band has. |
Click the Table for Larger View
To see the original image, click images
Question 2: Image Filteriing by Moduler
[Part 1] Working with spatial filters
* There are various filtering algorithm in PCI Geometica. We can find filtering function in Module Librarian. Through Moduler, I could find 27 ways of filtering.
*Directory: Algorithm --> Image Processing --> Image Filtering
* Description:
Coherent signal scattering in SAR data often causes image speckles or salt and pepper effects. Speckling is inherent in most images and can inhibit accurate image interpretation. There are several image filters in Focus to help manage image speckling.
Coherent signal scattering in SAR data often causes image speckles or salt and pepper effects. Speckling is inherent in most images and can inhibit accurate image interpretation. There are several image filters in Focus to help manage image speckling.
Filters can enhance or subdue the details of an image. They can also be adjusted to sharpen, smooth, or detect hidden edges that are present in an image but not immediately visible. You can use the low-pass and high-pass filters to reduce graininess and highlight edge details in images. There are also specialized filters that you can use to reduce sensor noise and to clean up radar imagery.
Filter computations are based on pixel samples drawn from a moving sample set, referred to as the kernel. The Kernel samples the image pixels and applies the filter to the center pixel in the sample. Once the filter is applied to the first sample, the kernel moves one pixel to the right and re-applies the filter until the entire image has been sampled. The kernel dimensions, measured in pixels, must always be an odd number; for example, 3x3 or 11x15. When the entire image has been sampled, Focus applies the changes to the entire image in the view pane. (Geometica Help)
Filter computations are based on pixel samples drawn from a moving sample set, referred to as the kernel. The Kernel samples the image pixels and applies the filter to the center pixel in the sample. Once the filter is applied to the first sample, the kernel moves one pixel to the right and re-applies the filter until the entire image has been sampled. The kernel dimensions, measured in pixels, must always be an odd number; for example, 3x3 or 11x15. When the entire image has been sampled, Focus applies the changes to the entire image in the view pane. (Geometica Help)
* Personal Findings:
By experiencing the image filtering function on band 1, band 2 and band 3. I found that each filter perform differently depend on different types of noises. So, I assume that specific filters may have significant impacts on particular noises and artifacts. Therefore, through Moduler, I performed several filter functions on each band.
By experiencing the image filtering function on band 1, band 2 and band 3. I found that each filter perform differently depend on different types of noises. So, I assume that specific filters may have significant impacts on particular noises and artifacts. Therefore, through Moduler, I performed several filter functions on each band.
Question 2: Image Filteriing by PCI focus
By using PCI focus, we are also able to operate filtering function. The reason why I tried to make filtered image with this operation is that I wanted to see what filtering function can eliminate dots in band 3. Using this way is useful because it has the "Apply to image" function to see the filtering result immediately.
Through Moduler, it was time-consuming to try all filtering operations to erase one specific noise. I tried Gaussian Filter and Average (Mean) Filter to eliminate dots all over the image, however, both filtering operation was not able to eliminate all dots on Band 3. However, with Mode in Low Pass Filtering in PCI Focus, I was able to see that this filtering system can clearly eliminate dots in Band 3.
Question 3: Image Enhancement
[Part 1]Comparing non-Filtered image and Filtered image
Analysis:
- By filtration, I was able to enhance an true colour image with less noise. When I compare those two images above, filtered true colour image has less noise with clear view.
- The original true colour image has red dots all over the image that TM3 (red band) was containing. Moreover, I am able to see a long thin stripe on the top that TM 2 had. The white band that TM 1,2,3 had can be clearly seen in the Original true colour image.
- On the other hand, the second true colour image that I create after filter operation has less noises. I operate all different and the best filtering system on each band. I used, FMO for TM1, FME for TM2, and MNFNR for TM3. (Too see the results, see Question 2) This filtered True Colour image does not have the thin stripe on the top of the image. Also,
By Using Maximum Noise Fraction Base Noise Removal (MNFNR), which is another way of removing bands affected by noise in a dataset, Dots all over the image (made by Band 3) are completely eliminated. Moreover, Even though I can still see the white band on the bottom-left of the image, the white band is much lighter than one in the original image.
By Using Maximum Noise Fraction Base Noise Removal (MNFNR), which is another way of removing bands affected by noise in a dataset, Dots all over the image (made by Band 3) are completely eliminated. Moreover, Even though I can still see the white band on the bottom-left of the image, the white band is much lighter than one in the original image.
Analysis:
- As those two false colour images above shows, the original false colour image and the filtered false colour image are very different. I these are false colour images that has NIR band as red colour, TM3 (Red) as Green colour, and TM2 (Green) as Blue colour.
- The original false colour image also has red dots all over the image that TM3 (red band) was containing. Moreover, I am able to see a long thin stripe on the top that TM 2 had. The white band that TM 1,2,3 had can be clearly seen in the Original false colour image.
- On the other hand, the second false colour image that I create after filter operation has less noises. Dots
On the other hand, the second true colour image that I create after filter operation has less noises. In order to get the filtered & combined image, I operate all different and the best filtering system on each band. The same as True Colour image, I used, FMO for TM1, FME for TM2, and MNFNR for TM3. (Too see the results, see Question 2). In the result image, there was no the thin stripe on the top of the image. Moreover, Maximum Noise Fraction Base Noise Removal (MNFNR), which is another way of removing bands affected by noise in a dataset, nicely worked well to remove little red dots in Band 3. Also, unlikely true colour image, the white band on the bottom-left is barely seen.
Question 3:True-colour and False Colour image with Enhanced image
[Part 1] True Colour Image
 |
| a-1. Filtered True Colour Image |
 |
| b-1. True Colour Enhanced map with Scaling Factor 2 |
Methods:
The first true colour map is a filtered true colour map (Band 1 is filtered by FMO, Band 2 is filtered by FME and Band 3 is filtered by MNFNR). The second Map is made the same as the first map except setting band 3 with enhanced band with scaling factor = 2.
a-1. RED=filtered band 3 by MNFNR
GREEN=filtered band 2 by FME
BLUE=filtered Band 3 by FMO
b-1. RED=Enhanced band with S.F 2,
GREEN=filtered band 2 by FME
BLUE=filtered Band 1 by FMO
[Part 2] False Colour Image
 |
| a-2. Filtered False Colour Map |
 |
| b-2. False Colour Enhanced map with Scaling factor 2 |
Methods:
The first true colour map is a filtered False colour map (Band 2 is filtered by FME and Band 3 is filtered by MNFNR and the original NIR (TM 4) is used). In the second Map, instead of using MNFNR filtered band 3, I used enhanced band with scaling factor = 2. a-2. RED=NIR,
GREEN=filtered band 3 by MNFNR,
BLUE=filtered Band 2 by FME
b-2. RED=NIR,
GREEN=Enhanced Band with S.F 2
BLUE=filtered Band 2 by FME
Question 4: Road Extraction
[Part 1] Getting a raster road map
Method
- Go to “Tools” to open “Algorithm Librarian”
- Click the “Find” button and type “THR” (Thresholding Image to Bitmap) and open the algorithm;
- Enter the threshold minimum (58) and maximum (160) pixel value
- “Run” the algorithm.
Result
 |
Road Extraction image with Threshold Min: 58, Threshold Max: 160 |
Analysis
- Thresholding is the easiest or simplest method of creating a binary layer to extract the roads from the rest of the image. Different threshold minimum and maximum number changes clearness of roads and the amount of built-up area. Although we are able to extract the roads, there are needs to clean up the classes that has pixel value falling into threshold 58 to 160. (Such as built-up area and other land cover types will be included). A more elaborate and concise tool may be used.
Methods:
Within the raster format, we will be able to edit delete spurious noise and add more road if needed. In order to digitize the map by hands, we might need to have true colour map for comparison.
Analysis:
This method is time consuming; however, by doing this work, we can get more clear road feature in the image.
[Part 3] How to Make the Road-line Clear
Methods:
[Part 2] A map after eliminating spurious noise
 | |
Within the raster format, we will be able to edit delete spurious noise and add more road if needed. In order to digitize the map by hands, we might need to have true colour map for comparison.
Analysis:
This method is time consuming; however, by doing this work, we can get more clear road feature in the image.
[Part 3] How to Make the Road-line Clear
Methods:
From the Algorithm Librarian, Find BIT2Line This module converts a bitmap layer to a line layer.
Analysis:
This Algorithm make road line more clear. It converted image type from raster to vector, and this module converts a bitmap layer to a line layer. By operating this algorithm, little noises that are not caught as a line are eliminated, so this module clears small noises that human made during digitizing.
[Part 4] Overlaying with True Colour Image
Analysis:
The true colour map, which is already filtered, can be overlayed with the extracted road map. This tells us road feature of the area.
Subscribe to:
Comments (Atom)