Feature description Flashcards
what are the four types of feature description
Statistical Features
Contour features
Moments
Texture features
what is feature extraction
generally, it is a process that is performed after segmentation, that allow to extract specific features from the image
what are the statistical features
are the features that are extracted from the distribution of labels in the image
what is the Normalized Difference Vegetation Index
we really NDVI index where we rely on the near-infra red-spactrum which is not visible to the human eye but very close to the infrared spectrum
what is the contour feature
allows providing information about the shape of the object
why we maynot want to use edge detector for the contour
usually for the contour features we more interest in a sequence, for this we may want to use chain code algorithm
explain the chain code algorithm
uses a certain pattern for the order of pixels ;
initially look for the first pixel in the object .
iteratively compute the direction using mod7(n+5)
until we reach the goal location
what are some exmaples of the contour use case
it allows the computation of the area and the centroid
what is a polygonal approximation
the idea consists of describing the object using shapes
How can we perform a polygonal approximation
for a given object. we initially draw the long line linking the two sides of the object, then we iteratively compute the maximum available error, to later link it and obtain a polygon each time
what are some shape descriptors
The form factor , compare the opbject perimeter with its circle perimeter
convexity: makes the relation between the convex perimeter and the original object perimeter
Aspect ration: the maximum and the minimum diameter
explain the Fourier contour description
The idea is the we take the distance to the centroid as the function of the angle , we would obtain a periodic function describing signal we can then use the Fourier transform
how t deal with different r(θ) in a shape
we can use the curvilinear coordinates to express the radius with respect to the coordinates
what is the fractal analysis
we performs some sort of measurement to compute the scale in terms of the length,to later on obtain the line and compute the corresponding angle winch reflect the fractal value
what is object moment
the moment equation describes the double summation on every axe for the pixel, know that
for p=0,q=0 it simply corresponds to the area of the shape
ΣxΣy x^q y^q f(x,y) to eliminate the influence of the shape position, the mean is subtracted
ΣxΣy (x-meanx)^q (y-mean_y)^q f(x,y)
then we introduce normalization to allow making it scale-invariant and translation invariant (normalized μ00 )
it is still not rotation invariant,
finally, we introduce the 7 Hu invariant to make rotation invariant
