Lecture 5: Fundamentals of Image Processing Flashcards
What captures a lot of information?
Images
What are the typical sizes?
320x240
640x480
1280x720
What size is QVGA?
320 x 240
What size is VGA?
640 x 480
What size is HD?
1280 x 720
What is filtering?
Certain components are accepted or rejected.
What does a low-pass filter do?
It smooths and image(allows low frequency)->blurring(smoothing) effect on an image used to reduce image noise.
What does a high-pass filter do?
It retains the contours(also called edges) of an image(high frequency)
What is the motivation of Low-Pass filtering?
noise reduction
What is salt and pepper noise? What is it also known as?
It presents itself as sparsely occurring white and black pixels sometimes seen on digital images.
It is also known as impulse noise.
What is gaussian noise?
It has variations in intensity drawn from a Gaussian normal distribution.
What are 2 example of Low-pass filtering?
-Salt and pepper noise
-Gaussian noise
How do you move average in 1D?
You replace each pixel with an average of all the values in its neighborhood.
How do you move average in 2D?
You replace each pixel with an average of all the values in its neighborhood.
What is the motivation of High-Pass filtering?
edge detection, an idealized line drawing
What do edge contours in the image correspond to ?
important scene contours
What are Edges?
Edges are sharp intensity changes.
How can edges be represented as functions?
f(x,y)
How can edges look like?
steep cliffs
An edge is a place of what?
a place of rapid change in the image intensity function
What is a fact about images?
They are noisy.
What are 4 examples of image noises?
Light Fluctuations
Sensor Noise
Quantization effects: from input values to quantized values
Finite Precision
When it comes to the summary on Smoothing Filters, what values does it have?
positive values
When it comes to the summary on Smoothing Filters, what does the “low-pass” filter remove?
It removes “high-frequency” components.
When it comes to the summary on Derivative Filters, what signs does it have and what are they used for?
It has opposite signs that are used to get high response in regions of high contact.
When it comes to the summary on Derivative Filters, what does the “high-pass” filter highlight?
It highlights “high-frequency” components
What do you do in non-maximal suppression?
You identify local maxima along a gradient direction.
What’s involved in the Canny edge-detection algorithm?
-computing gradient of smoothed image in both directions
-discarding pixels whose gradient magnitude is below a certain threshold
- non-maximal suppresion
What is the process of the Canny edge-detection algorithm?
-Take a grayscale image. If not grayscale, convert it into a grayscale by replacing each pixel by the mean value of its R,G,B components.
-Convolve the image with x and y derivatives of Gaussian filter
-Threshold it(set to 0 when the pixel value is below a given threshold)
-Take local maximum along gradient direction.
What is Thinning also known as?
Non-maxima suppression, local-maxima detection along edge direction
How do you build a panorama? 4 steps
-match/align images
-detect feature points in both images
-find corresponding pairs
-use these pairs to align images
What is problem 1 when it comes to point features?
Detecting the same points independently in both images, if they are in the field of view.
There is no chance to match!
What do you need to solve problem 1?
a repeatable feature detector
What is problem 2 when it comes to point features?
For each point, identify its correspondence in the other images
What do you need to solve problem 2?
a reliable and distinctive feature detector
What’s true with some patches?
Some patches can be localized or matched with higher accuracy than others
Information can be what two things?
Useful or Redundant
How do you get useful information from images?
Selecting features that:
are distinctive
do not vary much in appearance
can be detected & matched very fast
What’s an example of useful information?
using point features like corners
What is the property of corner detection?
shifting a window in any direction should give a large change of intensity in at least 2 directions
When is corner detection identified as a “flat” region?
When there is no intensity change
When is corner detection identified as an “edge”?
when there is no change along the edge direction
When is corner detection identified as a “corner”?
when there is significant change in at least 2 directions.
