aula_02 Flashcards
What are the 2 types of low level image processing?
-Arithmetic operations
-Pixel relations
What kind of Arithmetic operations are there?
– Addition
– Substraction
– Multiplication
– Division
as well as logical operations like :
AND, OR, NOT, …..
What many image processing operations make use ?
Of spatial relationships between pixels.
A number of methods have been devised to specify pixel
neighbours and calculate distance.
* The 4-neighbours of a pixel (x,y) are the closest pixels in
horizontal and vertical directions (D4) - like a cross
* The 8-neighbors are the 4-neighbors plus the four
closest pixels in diagonal direction (D8) - all the neighbours
How do we calculate the pixel distance?
it can be calculated with the euclidean way but the sqrt makes it expensive to calculate, despite being the most accurate
–Euclidean (L2): 𝐷 = sqrt([(𝑥 − 𝑢) 2 + (𝑦 − 𝑣) 2 ])
– City-block (L1): D = |𝑥 − 𝑢| +| 𝑦 − 𝑣|
– Chessboard (Linf): D = max (|𝑥 − 𝑢| ,| 𝑦 − 𝑣|)
What is intensity level slicing?
It means highlighting a specific range of
intensities in an image.
Is threshold a specific case of intensity level slicing?
Yes.
What types of threshold there are?
There are:
-Threshold binary
-Threshold binary
-Threshold binary
-Threshold binary
How many types of threshold there are?
There are 5 types:
-Threshold binary
-Threshold binary, inverted
-Truncate
-Threshold to zero
-Threshold to zero, inverted
What is threshold binary?
it corresponds to the maxval accepted ( normally 1) if the value of the function src(x,y) is > tresh value, and it is 0 otherwise
What is threshold binary inverted?
It is the same as threshold binary but it is 0 if the value of the function src(x,y) is > tresh value and it is 1 or max value otherwise
What is Truncate?
If the function value is > thresh value, it truncates it to the thresh value, otherwise it is the value of the function itself src(x,y)
What is Threshold to Zero?
It is src(x,y) if the function > the thresh value and it is 0 otherwise
What is Threshold to Zero, Inverted?
It is src(x,y) if the function < the thresh value and it is 0 otherwise ( if the src(x,y) > thresh value)
What is an histogram? What they allow us to see?
- An histogram is a graphical display of
tabulated frequencies. - Typically represented as a bar chart
*histograms allow us to see the colour
or intensity distribution
Wow can the collected counts of data be organized
in a histogram?
Into a set of predefined bins.
What are the three important parts of the histograms?
Important parts of an histogram:
– dims: The number of parameters you want to collect data.
– bins: The number of subdivisions in each dim.
– range: The limits for the values to be measured.
What are the two parts of the Operation based on histogram
– Determination and analysis of histogram
– Image improvement based on the histogram.
How is an histogram ideal distributed?
An ideal histogram is evenly
distributed, edge to edge,
not up the sides.
And a histogram for a dark
subject?
it is just more shifted to the left to
represent the tones of the
subject. It is not wrong
And a histogram for a light
subject (e.g., a white cat)?
Since it mostly has light tones in the scene
and few dark areas, it is shifted to the right compared
to the dark subject
And how can it be a graph for a overexposed image?
The gap on the left side of the graph indicates there a lack of blacks in the image.
It also means you will lose lots of detail in the white areas
that may not be recoverable. In this case, shift
to give your image less exposure and shoot
the scene again.
And if the opposite happens? How do you call it?
This histogram shows the opposite. Now we see a gap on the
right side of the graph indicating there are no whites
represented, so the image will be dark – too dark. You can
safely give the image more exposure until you see the tones
just touch the right edge of the histogram.
It is called Underexposed
Again, what defines an underexposed image? (exposition)
Loss of details in dark areas
Again, what defines an overexposed image? (exposition)
Loss of details in bright areas
Histograms can also have different intensity levels? (intensity)
Yes, for example: 256, 64 and 6 intensity levels
And how is contrast defined? (contrast)
Contrast is defined as the difference in intensity between two objects in an image. If the contrast is too low, it is impossible to distinguish between two objects, and they are seen as a single object. A lot of zones that dont have values, whereas in other zones there are very high values
Can the contrast be manipulated? (contrast)
Yes.
Using for example contrast stretching and also histogram equalization (extension of contrast stretching)
How does the histogram equalization occurs?
-It is analysed by a function T(z) to reshape
the image to make its histogram flat and wide.
– A typical solution is to use the cumulative histogram
(integral of intensity histogram) as the intensity
mapping function.
Are there also color histograms?
Yes. They have a histogram for each of the colors (RGB for example)
What is image smoothing or blurring? (filtering)
It is also known as neighbourhood averaging:
– Used to remove noise or a pre-processing for edge
detection
– Filtering operation in which a weighted array is
moved over the original image while computing the
pixel as a weighted average:
Can it use different filters/kernels?
Yes.
For each pixel there is aplied a operation that allows a kernel to modify the value of the pixel itself and its neighbours
And how does it work on the borders?
Since it has no neighbours in the corners/borders, the kernel can aplly different values, like for example:
- constant value;
- value of the closest border;
- mirror at the image boundary;
- periodically repeat along axis
And what is the average filter for noise reduction?
– It is Kind of the low pass filter, the larger the size, the
larger the cut-off frequency.
– This is an approximation since smoothing operation
is performed in space and not in frequency domain.
What examples of linear filter are there?
-The box filter with nine neighbours with 1 value and the outsider neighbours with 0
-The gaussian filter that has 9 in the pixel, 5 in the vertical and horizontal neighbours and then starts to decrease
-The “Laplace” or “Mexican Hat” filter that has 16 value in the pixel and then negatives around and 0 in the outsider neighbours
And does the median filter works?
It apllies a sorting of the pixels values and then ????
And is there any other version of the median filter?
Yes, a weighted median filter that weight the sorted pixel values
How do geometrical transformations essentiatly occur?
The Value of image I at original location (x,y)
moves to new position (x′,y′) in transformed
image I′ based on a geometric mapping T
And what kind of geometrical transformations can be applied?
There are Linear (Afiine and Projective) and Non linear types
Which ones are Linear?
In the Affine group:
– Translation, rotation, and scaling (contract and stretch) which does not preserve lines and parallelism
In the Projective:
– It is a Particular case of affine which does not preserve parralelism
Which ones are Non Linear?
For example the :
-twirl, ripple, sphere, …
And what are the Geometrical transformations used for?
They are used for example:
– Correction of distortion (Camera lens correction)
– Texture mapping (Computer Graphics)
– Morphing (Special effects in movies)
