L4 IMG PROCESS SUMMARY Flashcards
● What does point processing mean?
When you change the brightness on your TV, you change the brightness of each pixel in the same way. That is an example of point processing.
Point processing is an operation that calculates the new value of each pixel g(x,y) based on the value of the pixel in the same position f(x,y) and some arithmetic operation.
● Image histograms ‐ not the Matlab code but the definition, how does it look like, why is it useful, what means histogram stretching, without the formula
An image histogram is a plot of the relative frequency of occurrence of each of the permitted pixel values in the image against the values themselves.
For a grayscale image, the histogram can be constructed by simply counting how many times each gray value (0-255) occurs within the image.
The histogram is actually a bar graph. On the x-axis we have the range of intensity values within the image (from 0-255) and y-axis shows the number of times each value actually occurs within a particular image.
It can sometimes happen that the contrast (difference between dark and light) in a gray scale image is not satisfactory. This happens because the minimum and maximal nonzero values on a histogram are not 0-255 but some other values a>0 and b<255. This poor contrast can be improved by the so called histogram stretching, which is a simple image enhancement that tries to “stretch” the range of intensity values it contains to span a desired range of values, e.g. full range of pixel values that the image type concerned allows.
● Thresholding – what it is, why it is useful, what is the result?
Thresholding is an useful point processing operation. It transforms a gray scale image in a black and white (binary) one by choosing a gray level T in the original image and thenturning every pixel black or white according to whether its gray value is greater than or less than T.
● What means neighborhood processing ? Filtering: what does it mean? How does the mean filter works , median filter?
In neighborhood processing the value of a pixel in the output image is determined by the value of the pixel at the same position in the input and the neighbors together with a neighborhood processing operation. Neighborhood processing can be used for example to filter an image.
Filtering means to improve the image quality by reducing or even removing the noise. Mean filter: improve image quality by changing the intensity of the noisy pixel to the mean value of its neighbors including itself (sum). Median filter: replace the noisy pixel with the median value of its neighbors including itself. The median value of a group of numbers is found by ordering the numbers in increasing order and picking the middle value.
● What is convolution (correlation) and how does it work?
Correlation is another form of neighborhood processing which also works by applying a filter to each pixel in an image. With correlation, the filter is called kernel and plays a more active role.
- first of all the kernel is filled by numbers (kernel coefficient) that are not always equal to 1 like we have seen with filters
- imagine the kernel as a magnifying glass that scans over the whole image centers on each pixel and changes its value
- suppose we correlate the image f(x,y) with the kernel h(x,y) and the result is a new image g(x,y)
- the kernel coefficient weigh the pixel value they are covering and the output of the correlation is a sum of weighted pixel values
this output value will be written in the pixel g(x,y)
● Edge detection. You do not have to memorize the kernels, only to know how they work.
Edge detection: an edge is a local discontinuity in the pixel value that exceeds a given threshold. Edge detection consists of creating a binary image from a gray scale image where non-background pixel values correspond to object boundaries. Edge detection can be obtained by a correlation with a kernel.
● What is a morphological operation? Give one example. Erosion, dilation, closing opening. You do not need to memorize the algorithm, but you should know where do we need them.
Morphology operates like the other neighborhood processing methods by applying a kernel to each pixel in the input image. The most important morphological operators are dilation and erosion. All other operations can be defined in terms of these primitive operators.
Morphological operators take a binary image and a structuring element as input and combine them using a set operator (intersection, union, inclusion, complement). If we have an image A and a structuring element B then we speak of the erosion/dilation of A by B.
To perform erosion of a binary image we center the STREL on each ‘1’ pixel in the image. If any of the neighborhood pixel is ‘0’ then the pixel in question is switched to ‘0’ (decreased in size). Erosion might be useful for example if we want to count some coins. In general, erosion of an image results in objects becoming smaller.
To perform dilation we center the STREL on each ‘0’ pixel in the image. If any of the neighborhood pixels is ‘1’ then the pixel in question is switched to ‘1’. In general, dilating an image results in objects becoming bigger, small holes being filled and object being merged.
Closing is the name given to the morphological operation of dilation followed by erosion with the same structuring element
Opening is the morphological operation of erosion followed by dilation with the same structuring element.
● What is a BLOB? What means connectivity and what kinds of connectivity we usually have? What means connected component labeling? What do we get as result of this labeling?
BLOB stands for Binary Large OBject and refers to a group of connected pixels in a binary image, also called connected components. The term “large” indicates that only objects of a certain size are of interest and that “small” binary objects are usually noise.
A number of different algorithms exist for finding the BLOBs and such algorithms are usually referred to as connected component analysis or connected component labeling. These algorithms search for connected pixels and assigns a label to each found BLOB. Background is usually 0 and then the first BLOB gets the label 1., the second 2 etc.
Extracted connected components is called labeling. Each connected component is given a label
