Image Processing at University Of South Florida | Flashcards & Summaries

Suggested languages for you:

# Lernmaterialien für Image Processing an der University of South Florida

Greife auf kostenlose Karteikarten, Zusammenfassungen, Übungsaufgaben und Altklausuren für deinen Image Processing Kurs an der University of South Florida zu.

TESTE DEIN WISSEN
Lösung anzeigen
TESTE DEIN WISSEN
• Consider each pixel (i,j)
• Calculate dispersion for all possible rotations about pixel
• Choose the minimum dispersion
• Assign the pixel the average brightness in the chosen mask
Lösung ausblenden
TESTE DEIN WISSEN
What is the H in HSI and what is it’s range?
Lösung anzeigen
TESTE DEIN WISSEN
It describes Hue, or the color itself, represented by an angle from 0-360
• 0 is red
• 120 is green
• 240 is blue
Lösung ausblenden
TESTE DEIN WISSEN
Averaging with limited data validity
Lösung anzeigen
TESTE DEIN WISSEN
Methods that average with limited input data validity try to avoid blurring by averaging only those pixels which satisfy some criterion, the aim being to prevent using pixels that are invalid or very noisy

• Simple criterion is to define a brightness interval of invalid data — only pixels with valid gray levels are utilized in averaging
• Only values of pixels with invalid gray levels are replaced with an average of their neighborhoods
• Only valid data contributes to the averages
Lösung ausblenden
TESTE DEIN WISSEN
Median Filtering
Lösung anzeigen
TESTE DEIN WISSEN
• The median divides the higher half of a probability distribution from the lower half
• Median filtering is non linear
• Reduced impulse noise quite well because it is not affected by individual noise spikes

Some useful notes:
• Nonlinear/non-separable
• Good for removing isolated noise pixels
• Preserves edges
• Problem with corners
• Creates “flats”

Complexity: O(N^2M^2 logM^2)
The log happens because of the sorting that must take place to find the median

Can be implemented incrementally
• Sort the the window and create list of pixels
• Move window by one pixel
• Remove M pixels from the sorted list O(M^2)
• Insert M pixels into the sorted list O(M^2)
• Hence O(N^2M^2)
• Use histogram arrays for incremental representation
Lösung ausblenden
TESTE DEIN WISSEN
Lösung anzeigen
TESTE DEIN WISSEN
Segmentation using variable thresholds. Threshold values vary over the image as a function of local image characteristics. Local thresholds are position dependent.

Ways of calculating thresholds:
• Mean
• Median
• (Max+min)/2
• K1 * local mean + k2 * standard dev

You need a large window size to cover sufficient foreground and background pixels otherwise a poor threshold is chosen.

Band thresholding:
Saying a pixel is foreground if it belongs to D, a set of gray levels, otherwise it’s background.
Semi thresholding:
Foreground if it’s greater than a certain value, background otherwise.

Lösung ausblenden
TESTE DEIN WISSEN
Bimodal Histogram
Lösung anzeigen
TESTE DEIN WISSEN
If an image consists of approximately the same gray-level that differ from gray level of background. Essentially two major groups within the image — foreground and background. Bimodality does not determine correct threshold segmentation.
Lösung ausblenden
TESTE DEIN WISSEN
Otsu’s Algorithm
Lösung anzeigen
TESTE DEIN WISSEN
Popular approach to automatic threshold detection. Underlying idea is to test each possible threshold and compute gray-level variances of both foreground and background.
When weighted sum of variances is minimal we can deduce that the threshold is separating the histogram in some “best” sense.
Lösung ausblenden
TESTE DEIN WISSEN
Optimal Thresholding
Lösung anzeigen
TESTE DEIN WISSEN
Alternative approach that seeks to model histogram of an image using weighted sum of two or more probability densities with normal distribution.

Algorithm:
• Select the median value of the original image
• 1/2 (expected background pixel + expected foreground pixel)
• (Loop) divide all pixels into two subsets — pixels less than and greater to threshold
• Find the average of the teo new images
• Calculate the new threshold by averaging the two means
• If the difference between the previous threshold value and new threshold value are believe a specified limit, you’re finished
Lösung ausblenden
TESTE DEIN WISSEN
Brightness Transformation
Lösung anzeigen
TESTE DEIN WISSEN
Dependent on properties of the pixel itself. There are two main classes of pixel brightness transformation
Brightness corrections
• Pixel position dependent
• Modifies pixel brightness taking into account its original brightness and position in the image

Gray-scale transformations
• Pixel position independent
• Changes brightness without regard to position in the image
Lösung ausblenden
TESTE DEIN WISSEN
Position-dependent brightness correction
Lösung anzeigen
TESTE DEIN WISSEN
Sensitivity of image acquisition and digitization depends on position of the image. The goal is to fix image degradation—caused by varying light levels in images. We use calibration with reference images.
Lösung ausblenden
TESTE DEIN WISSEN
Histogram modification
Lösung anzeigen
TESTE DEIN WISSEN
When plotted on a histogram, images with poor contrast tend to have a few large clusters in intensity values that dominate the image. The dynamic range of an image does not match the dynamic range of the display. Histogram stretching simply spreads the intensity values more evenly throughout the range scale.
Lösung ausblenden
TESTE DEIN WISSEN
Directional Smoothing
Lösung anzeigen
TESTE DEIN WISSEN
Taking vertical, horizontal, and diagonal 1D windows around the pixel and computing the mean of each. The one that is closest to the value of the pixel being tested is selected
Lösung ausblenden
• 3624 Karteikarten
• 99 Studierende
• 0 Lernmaterialien

## Beispielhafte Karteikarten für deinen Image Processing Kurs an der University of South Florida - von Kommilitonen auf StudySmarter erstellt!

Q:
A:
• Consider each pixel (i,j)
• Calculate dispersion for all possible rotations about pixel
• Choose the minimum dispersion
• Assign the pixel the average brightness in the chosen mask
Q:
What is the H in HSI and what is it’s range?
A:
It describes Hue, or the color itself, represented by an angle from 0-360
• 0 is red
• 120 is green
• 240 is blue
Q:
Averaging with limited data validity
A:
Methods that average with limited input data validity try to avoid blurring by averaging only those pixels which satisfy some criterion, the aim being to prevent using pixels that are invalid or very noisy

• Simple criterion is to define a brightness interval of invalid data — only pixels with valid gray levels are utilized in averaging
• Only values of pixels with invalid gray levels are replaced with an average of their neighborhoods
• Only valid data contributes to the averages
Q:
Median Filtering
A:
• The median divides the higher half of a probability distribution from the lower half
• Median filtering is non linear
• Reduced impulse noise quite well because it is not affected by individual noise spikes

Some useful notes:
• Nonlinear/non-separable
• Good for removing isolated noise pixels
• Preserves edges
• Problem with corners
• Creates “flats”

Complexity: O(N^2M^2 logM^2)
The log happens because of the sorting that must take place to find the median

Can be implemented incrementally
• Sort the the window and create list of pixels
• Move window by one pixel
• Remove M pixels from the sorted list O(M^2)
• Insert M pixels into the sorted list O(M^2)
• Hence O(N^2M^2)
• Use histogram arrays for incremental representation
Q:
A:
Segmentation using variable thresholds. Threshold values vary over the image as a function of local image characteristics. Local thresholds are position dependent.

Ways of calculating thresholds:
• Mean
• Median
• (Max+min)/2
• K1 * local mean + k2 * standard dev

You need a large window size to cover sufficient foreground and background pixels otherwise a poor threshold is chosen.

Band thresholding:
Saying a pixel is foreground if it belongs to D, a set of gray levels, otherwise it’s background.
Semi thresholding:
Foreground if it’s greater than a certain value, background otherwise.

Q:
Bimodal Histogram
A:
If an image consists of approximately the same gray-level that differ from gray level of background. Essentially two major groups within the image — foreground and background. Bimodality does not determine correct threshold segmentation.
Q:
Otsu’s Algorithm
A:
Popular approach to automatic threshold detection. Underlying idea is to test each possible threshold and compute gray-level variances of both foreground and background.
When weighted sum of variances is minimal we can deduce that the threshold is separating the histogram in some “best” sense.
Q:
Optimal Thresholding
A:
Alternative approach that seeks to model histogram of an image using weighted sum of two or more probability densities with normal distribution.

Algorithm:
• Select the median value of the original image
• 1/2 (expected background pixel + expected foreground pixel)
• (Loop) divide all pixels into two subsets — pixels less than and greater to threshold
• Find the average of the teo new images
• Calculate the new threshold by averaging the two means
• If the difference between the previous threshold value and new threshold value are believe a specified limit, you’re finished
Q:
Brightness Transformation
A:
Dependent on properties of the pixel itself. There are two main classes of pixel brightness transformation
Brightness corrections
• Pixel position dependent
• Modifies pixel brightness taking into account its original brightness and position in the image

Gray-scale transformations
• Pixel position independent
• Changes brightness without regard to position in the image
Q:
Position-dependent brightness correction
A:
Sensitivity of image acquisition and digitization depends on position of the image. The goal is to fix image degradation—caused by varying light levels in images. We use calibration with reference images.
Q:
Histogram modification
A:
When plotted on a histogram, images with poor contrast tend to have a few large clusters in intensity values that dominate the image. The dynamic range of an image does not match the dynamic range of the display. Histogram stretching simply spreads the intensity values more evenly throughout the range scale.
Q:
Directional Smoothing
A:
Taking vertical, horizontal, and diagonal 1D windows around the pixel and computing the mean of each. The one that is closest to the value of the pixel being tested is selected

### Erstelle und finde Lernmaterialien auf StudySmarter.

Greife kostenlos auf tausende geteilte Karteikarten, Zusammenfassungen, Altklausuren und mehr zu.

## Das sind die beliebtesten Image Processing Kurse im gesamten StudySmarter Universum

##### Bioprocessing

Cape Peninsula University of Technology

##### Process of Teaching

University of Southern Mindanao