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.

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)

* 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.

[Part 2] Various Filter Module

[Band 1]

As I explained in the Question 1, several noises are founded in Band 1.

[Mode Filter(FMO)]

- Details: performs mode filtering on image data. The mode filter is used primarily to clean thematic maps for presentation purposes. The mode filter works with the mode of the gray-level values in the filter window that surrounds each pixel. Mode filtering is ideal for cleaning thematic maps for presentation purposes because it replaces small 'island' themes with larger surrounding themes. A mode filter is applied to a pixel only if the number of occurrences of the center pixel value in the 3x3 window that surrounds it is less than 3. The reason why we sett the center pixel value small is that it lead thin classes (such as streams or roads) are not removed when cleaning thematic maps.

- Outcome: FMO was not effective to get rid of noise in Band 1, still a white band on bottom-left of the image. Moreover, coherent Noise is still appearing near the lake.

[Gaussian filtering(FGA) ]

-Details: This filtering system can be used as a low-pass filter to blur the image, if only one Gaussian filter parameter SigmLow is specified. It also can be used as a band-pass filter to detect sudden intensity change in the image, if both SigmLow and SigmHigh parameters are specified. (In this experiment, I used this as a low-pass filter)

-Outcome: This Filtered outcome was worse than the original band. Gaussian Filter was not effective to eliminate coherent noise in Band 1 of the satellite image.

[ Sobel Edge Filter (FSOBEL) ]

- Details: FSOBEL performs Sobel edge detector filtering on image data. The Sobel edge detector filter creates an image where edges (sharp changes in gray-level values) are shown. Sobel edge detection produces an image where higher gray-level values indicate the presence of an edge between two objects. The Sobel edge detector filter computes the root mean square of two 3X3 templates.

- Outcome: This Filter operator is good to erase noises that I had in the Band 1. Through this filter, the white band on the bottom-left is eliminated, and the white hazy edge on the lake is also erased. 

However, some details of this image are lost by this filter, for example, it can not show difference between vegetation and water areas. That is because This Sobel edge detetor filter creates an image sharper in certain value. In this case, roads is emphasized with bright white colour and sharp edges.

 

[BAND 2]

[Filtered image by Median Filter (FME) ]

Details: The median filter computes the median value of the grey-level values within a rectangular filter window surrounding each pixel. This has the effect of smoothing the image and preserving edges. The dimensions of the filter window must be odd. The minimum filter size is 1 by 3 and the maximum filter size allowed is 33 by 33. The filter window does not need to be square. A bitmap specifies the area within the input layer which will be processed. Only this area will be filtered and the rest of the image will be unchanged. If no bitmap is connected, the entire database is processed.

Outcome: The filtered image by FME was very effective. It perfectly removed the black line on the top of the image and it also erased the white band on the bottom-left in Band 2. When I tried other filtering system, I found that these filtering system makes the black line on the top of the image even thicker and darker.

 

[BAND 3]

[Gaussian filtering(FGA) ]

-Details: This filtering system can be used as a low-pass filter to blur the image, if only one Gaussian filter parameter SigmLowSigmLow and SigmHigh parameters are specified. (In this experiment, I used this as a low-pass filter)
-Outcome: The filtered outcome by FGA was worse than the original band 3. Gaussian Filter was not effective to eliminate coherent noise in Band 3 of the satellite image. Compare to the original image, the lake's northern edge has much more hazy effect after operating this filtering function.

                                   

 [Averaging (mean) Filter (FAV)]

- Details: FAV performs average filtering on image data. The averaging (mean) filter smooths image data and eliminates noise. The average filter computes the mean (average) of the gray-level values within a rectangular filter window that surrounds each pixel.

The minimum filter size is 1x3 (or 3x1), and the maximum filter size is 1001x1001. The filter window need not be square. If a single value is specified, this value points to a bitmap segment that defines the area that is filtered. When four values are specified, these values define the x,y offsets and x,y dimensions of the rectangular window in the image that is filtered. 

- Outcome: The outcome of the FAV is used as Fuzzy image. This outcome as a filtering operation can be said to be ineffective. That is because, although all dots all over the image were in somehow reduced, but they were not completely eliminated. Furthermore, there was no elimination of the white strip on the bottom-left of the image. 

[Maximum Noise Fraction Based Noise Removal  (MNFNR)]

- Details:  MNFNR attempts to remove noise in a single image band using a procedure based on the maximum noise fraction transform.

Program MNFNR need a set of image bands one band is considered to have significantly more noise than the others, and it is desirable to transform that band such that its noise content is close to that of the other band. 

- Outcome: This Filtering is not in the same category in Algorithm Library. However this filtering operation created a the best filtration of TM3. While other two filtration failed to erase dots on all over the image, MNFNR clearly eliminated them. Furthermore, there was  elimination of the white strip on the bottom-left of the image as well. Thus, it can be said that MNFNR was succeful to clear all noises in TM3. 

To see Screenshots Click here