Week 2 - Image Formation & Feature Representation Flashcards
(40 cards)
What does the term Principal or Optical axis mean in regards to pinhole cameras?
The Z-axis is often referred to with these terms, and it assumes that the camera orientation is aligned with the Z-axis.
What is the point, where the Z-axis hits the image plane, called?
The principal point
How is the principal point located?
It is located at (0, 0, -f) from origin, where f is the focal length.
How does the Focal Length relate to Field of View?
As focal length increases, field of view decreases i.e. zooming in on something is an example of this.
As focal length decreases, field of view increases i.e. zooming out on a scene
How do you work out the coordinates of the projection on the image plane mathematically?
x = f(X) / Z
y = f(Y) / Z
Where X, Y, Z is the point on the object, and x, y is the same point on the image plane.
What happens to parallel lines when they ‘head towards’ the horizon?
Parallel lines will eventually appear to converge on a vanishing point
What are some problems with pinhole cameras?
- Pinhole size (aperture) must be ‘very small’ to obtain a clear image. However, if pinhole size is made smaller, then less light is received by the image plane.
- If pinhole is comparable to wavelength of incoming light, then diffraction effects blur the image
How do you ensure that the pinhole camera captures the sharpest image possible?
Diameter of pinhole camera = 2 square root of (f * wavelength of light)
Example:
If f = 50mm, and wavelength = 600nm, then diameter = ~0.35mm
What are some advantages in using the pinhole camera?
- Simple to understand
- Infinite depth of field
- No lens distortion
What are lenses used for?
Lenses are used to avoid the problems of using the pinhole camera, by capturing more light from the image plane, but retaining the same projection.
What does the term ‘f’ stand for in terms of lenses?
f = Focal length of the lens, which determines the lens’s ability to bend/refract light
How is intensity measured numerically?
Intensity = 0 if pixel is black
Intensity = 255 if pixel is white in an 8-bit image
What factors affect the colour of a pixel in an image?
Light sources:
- Emittance spectrum
- Geometry
- Directional attenuation
Objects’ surface properties:
- Reflectance spectrum
- Geometry
- Absorption
What are some typical use cases for Image Feature Representation?
- Image alignment
- 3D reconstruction
- Motion tracking
- Object/face recognition
- Indexing and database retrieval
What are image features?
A feature is a measurable property that describes the characteristics of an image or a region of images
How are image features often represented?
Often represented by scalars, vectors, matrices or tensors.
What factors make image matching hard to perform?
- Change in lighting
- Change in viewpoint
- Occlusions
- Partial matching
- Change over time
What are image regions & patches?
Image regions & patches are segments or rectangular image patches that are used to collect a wider area of information from an image
What is a feature vector?
A feature vector defines a set of descriptive features and concatenates them to produce a feature vector
What is the idea behind using feature vectors?
The idea is to remove redundant or irrelevant data
What makes histograms a good representation of colour?
- Invariant to translation and rotation
- Change slowly as viewing direction changes
- Change slowly with object size
- Change slowly with occlusion
What are texture features?
Texture features measure the frequency with which patterns of colour/grey levels appear
What are gradient-based features?
They are areas in an image which typically indicate boundaries of objects due to a spike in intensity i.e. intensity gradient.
How do you estimate gradients using spatial filtering?
You take your source pixel, and the area surrounding it, and apply a convolution kernel on top of the area. You then multiply the value at each pixel corresponding with the value in the kernel, before summing all the results together to form the final source pixel value.
