Computer Vision
Created @September 6, 2021 5:14 PM
Class
Type
Materials
Lecture 1
introduction
basic idea: make a computer see and understand images (computer is linked to sensors or camera('s))
We can not just copy human vision: because we don’t yet have a sufficient understanding of how our
visual system works. Also the architecture is different: brain is slow but parallel, Computer is fast but
mainly serial
search for surface (normal)
Lecture 2
H1-2: camera model and image formatting
three levels of vision processing
Low-level vision: image processing, denoising, filtering, image restoration, low-level feature
extraction
Mid-level vision: basic image analysis, segmentation, contour extraction, perceptual organization,
2-1/2D representation, 3D information recovery
High-level vision: understanding, detection and recognition. semantic labelling, activity, event
detection
pinhole camera
similar mechanics as eyes, lens etc.
a barrier (aperture/shutter) needed to block the light/rays. Because the light from the object is
shattered to every direction
→ reducing blurring, and
→ to much light on film causes burning on the film
Computer Vision 1
, pupil is constricted in daylight, because there is a lot of energy/light (so many lines in the picture)
and we only need a small amount. But in the night there is less energy so we need a wider
opening of the barrier/pupil to get more energy
smaller number for lens (f/2.8) means larger opening
adding a lens to capture more light. Called zoom lens. However can not always focus fully (purple
one) → reshape of lens needed (eye does this automatically, and new technology called liquid
lens is discovered).
geometric image formation
world coordinates to image coordinates
homogeneous coordinates: with this we can get away with add → the equation will not get any
longer with only the product (bec the vectors will be same size). the smaller the w (distance image
to pinhole, now in vector at the end) → smaller image
Computer Vision 2
,Computer Vision 3
, Intrinsic changes
geometric center is in the middle, but the center of an image is in the left upper corner, so we
need extra values (u0 &v0 ) to move the image. minus by the Y → Y now goes down instead
of up
Computer Vision 4
Created @September 6, 2021 5:14 PM
Class
Type
Materials
Lecture 1
introduction
basic idea: make a computer see and understand images (computer is linked to sensors or camera('s))
We can not just copy human vision: because we don’t yet have a sufficient understanding of how our
visual system works. Also the architecture is different: brain is slow but parallel, Computer is fast but
mainly serial
search for surface (normal)
Lecture 2
H1-2: camera model and image formatting
three levels of vision processing
Low-level vision: image processing, denoising, filtering, image restoration, low-level feature
extraction
Mid-level vision: basic image analysis, segmentation, contour extraction, perceptual organization,
2-1/2D representation, 3D information recovery
High-level vision: understanding, detection and recognition. semantic labelling, activity, event
detection
pinhole camera
similar mechanics as eyes, lens etc.
a barrier (aperture/shutter) needed to block the light/rays. Because the light from the object is
shattered to every direction
→ reducing blurring, and
→ to much light on film causes burning on the film
Computer Vision 1
, pupil is constricted in daylight, because there is a lot of energy/light (so many lines in the picture)
and we only need a small amount. But in the night there is less energy so we need a wider
opening of the barrier/pupil to get more energy
smaller number for lens (f/2.8) means larger opening
adding a lens to capture more light. Called zoom lens. However can not always focus fully (purple
one) → reshape of lens needed (eye does this automatically, and new technology called liquid
lens is discovered).
geometric image formation
world coordinates to image coordinates
homogeneous coordinates: with this we can get away with add → the equation will not get any
longer with only the product (bec the vectors will be same size). the smaller the w (distance image
to pinhole, now in vector at the end) → smaller image
Computer Vision 2
,Computer Vision 3
, Intrinsic changes
geometric center is in the middle, but the center of an image is in the left upper corner, so we
need extra values (u0 &v0 ) to move the image. minus by the Y → Y now goes down instead
of up
Computer Vision 4
