What is 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
What is the definition of an image histogram? How does it look like? Why is it useful?
An image histogram is a plot of the relative frequency of occurrence of each of the permitted pixel values in the image (y) against the values themselves (x)
Bimodal histogram = histogram has two prominent peaks where one peak corresponds to the background pixels and the other peak to the foreground pixel
Two distinctive peaks:
- Peak in lower range of brightness = dark background
- Peak in higher range of brightness = foreground objects (coins)
What does histogram stretching mean? (no formula)
Histogram stretching is a point operation when an image has a poor contrast (relevant intensity interval is much smaller than interval [0, 255]) Aka normalization or equalization
What does neighbourhood processing mean?
In neighborhood processing, the intensity value of each pixel in the output image g(x,y) is determine by the value of the pixel at the same position in the input image f(x,y) together with its neighbors and a neighborhood processing operation
What does filtering mean and how does the mean and median filter work?
Filtering is a type of neighborhood image processing used to reduce or even remove noise
- Applies to all pixels in the image, so good pixels might be worsened but noise reduced
A filter is a N x N array (image) with all elements (pixels) equal to 1
- Mean: averaging the values of the target pixel (0 or 255) and all its neighboring pixels
- Median: ordering the values of the target pixel and its neighbors in ascending order and assigning the middle value to the target pixel
How do kernels work?
Used in correlation/convolution as a filter: Kernel is like a magnifying glass that scans over the image f and alters the values of the pixel. When the scan is ready, a new image g of the same size is generated.
- Filled with kernel coefficients, numbers not necessarily equal to 1; Kernel coefficients are weighting the value of original pixels they are covering
- types: 3x3 mean filter, Gaussian blur kernel, Sobel horizontal edge kernel
What is convolution (correlation) and how does it work?
Correlation or convolution is a neighborhood processing technique that works by applying a filter to each pixel of an image, where the filter is called the kernel.
- Correlation between an input image f and a kernel h results in a new output image g
The kernal scans over the image and alters the values of the pixel so that a new image of the same size is generated.
Template matching (locate a certain object) and Edge detection (image pattern recognition) make use of correlation
What is a morphological operation? Give one example
Morphology is a branch of image processing which allows for the analysis of shapes in images. It works by applying a kernel, called structuring element(STREL) to each pixel in an input image.
Morphological operators take a binary image and a structuring element as input and combine them using a set operator (intersection, union, inclusion, or complement)
All other morphological operators rely on dilation and erosion, the two most important and primitive operators
Where do we need erosion, dilation, closing and opening? (no need to memorise algorithm)
Erosion: decreases the binary image’s object in size
- Center the STREL on each 1 pixel in the image
- If any neighborhood pixels is 0 then the pixel in question is switched to 0
Dilation: increases the binary image’s object in size
- Center the STREL on each 0 pixel in the image
- If any neighborhood pixels is 1 then the pixel in question is switched to 1
Closing: dilation then erosion (using the same structuring element) - To remove small holes and join narrow strips between objects
Opening: erosion then dilation (using the same structuring element) - To remove small, isolated noisy objects while main object preserves original size -Segmentation of objects of the same shape in an image
What is a BLOB?
A BLOB refers to a group of connected pixels in a binary image called connected components = Binary Large (only objects of a certain size are considered) OBject
What is connectivity and what kinds of connectivity do we usually have?
Connectivity determines whether or not two pixels are connected by recognizing which pixels are neighbors and which are no
- Most common types: 4-connectivity and 8-connectivity
- 8-connectivity is more accurate but 4-connectivity requires less computations (processes image faster)
What does connected component labelling mean? What do we get as a result of this labelling?
Connected Component Labelling is an algorithm to find BLOBs in an image by searching for connected pixels and assigning a label to each found BLOB. Usually, background is labelled 0, first BLOB is 1, second BLOB is 2, etc. The label can be translated into colour.
