Segmentation Flashcards
what are the four stage of image processing
image acquisition, when photons are converted into a matrix of numbers
pre-processing, used to correct image acquisition default or enhance image quality in order to facilitate…
the segmentation, that split the image into an object of interest and background, then each object of interest is described by,
feature extraction, that describes shape, colour, texture, etc
explain briefly the image segmentation
the segmentation is assumed to be be either a pixel belong to the object or the background in histogram segmentation is based on the idea of having range values such that the value among this range will be part of the image
Explain the histogram-based Segmentation
it is a type of segmentation where all the pixels are treated the same
Explain the image threshold , and the types of the threshold by the
the idea is simple such that everything below a certain value T is 1 otherwise it is 0 :
T can be fixed
T can be globally adaptive in situations where we do not know exactly what is the threshold and perform a global measurement
T can locally adaptive depends on the neighbourhood
what can we do in the case when we want to apply segmentation on an image that Is super noisy
initially performing directly segmentation, may lead to an unwanted results, so better omit the noise by using a median filter then normalize the image by the newly obtained one, to later on, apply the necessary threshold
why do we use the several type of color RGB
since our brain is only capable of differentiating 255 grey levels hence, if we encode the gray level using color our brain can differentiate between the lines that are close to each other, the blue to yellow convention is for the color blind people
what is the drawback of threshold
the drawback is deciding the threshold position is hard for the single curved image histogram
how to segment a color image
simply we can pick a threshold per color (R or G or B )
how to automatically choose a value for the threshold
we use what we call a percentile threshold, for instance, the galaxy image, if we know that 5 percent of the galaxies are covered by an image, hence we can rely on the cumulative histogram to put the threshold in the part that divide the distribution in 5 percent
what is the optimal threshold, and explain how it works
used to detect the most adequate threshold for the image,
Optimal threshold: put arbitrarily the threshold value, then separate the distribution according to the threshold, then for each part compute the centroid, then update the threshold, by putting it in the middle, keep on iterating till convergence , however note in a single curved may not work very well
is optimal threshold local or global
it is local since we rely on the initial threshold
explain the OTsu threshold
is defined by two classes foreground and background separated by a threshold, we have the within-class variance that defines how compact it is while between class variance aims to split the two variances as much as possible, then the separability is defined through optimizing the between-class variance which should as separate as possible and within-class should be as small as possible the way it is used is that we use the expression for every threshold value and we choose the optimum
is the otsu local or global threshold
it is a global one since we test every threshold
what is the entropy threshold
calculates the entropy below and above the threshold for each possible threshold value, then we put the threshold when the entropy is maximum
what is the multispectral threshold
it considers a segmentation on each image specter then merge them together
how to detect whether a specific region is blurred
generally blurred regions have low frequencies
How can we deal with images that are characterized by different regions
we can compute the local entropy according to a window, when the entropy is low means we have less variation and more blurr, while higher entropy defines the sharp parts , then we perform a threshold on the entropy image and a threshold on the origin one then we combine them
what is a border segmentation
it consists of a segmentation that is based on the information provided by borders, we usually enhance the border for instance using Sobel filter, so to separate the images, we say we are only interested in very sharp edge sobel >50
what is the downside of border segmentation
when the border are not very closed, this mislead the algorithm
what is region-based segmentation
it actually overcome the the downside of border segmentation
what is a region growing segmentation
generally, we start from the point in the object and we keep on recruiting neighbouring pixels, it is composed of two phases :
the split-phase: where the image is split recursively then image if it is not homogeneous ( generally we can use the variance or the entropy to check the homogeneity of the pixels ), we split the image to 4
finally the merge phase consist of merging all the parts such that if the two grouped regions satisfies the homogeneity criterion, merge the regions
explain the Model based segmenetation
Generally we know apriori the object we are looking for
explain the hough Transform
Aimes to detect line segment in an image,
the idea is to transform initially the line equation in terms of theta and ru: x cos(θ)+ysin(θ)=ρ where ρ is the distance to the origin point and θ is the angle with the x axis
so each point is represented by the above line equation, now we need to detect their intersection
we initially map those lines to point in in the ru the space and then the intersection corresponds to the point they intersect in In the same space
but before applying it we should initially apply the canny edge on the image
what is the intilligent scissors methods
it is a semi-automatics consisting of putting in the image then the algorithm tries to find the best path between these points
