Image formation
- refracted ray
- part of incident ray that enters a surface and is bent according to
Snell's law.
- Lensmaker's equation
- Relates the distance between the point being viewed and the lens to the
distance between the lens and the ideal image location.
- Optical power of a lens
- How strongly a lens bends light; measured in diopters
- Accommodation
- Adjustment of the focal length of a lens to bring different parts of a scene
into focus.
- Depth of field
- Range of distances in focus
- Chromatic abberation
- Prism effect of focusing light of different wavelengths at different
distances behind the lens.
- Fovea
- Area of highest concentration of cones in the visual system.
Segmentation by thresholding
- pixel
- individual grid element of a digital image
- intensity or grey level
- Measure of reflected light in a pixel
- image noise
- statistical variations in the observed grey levels; may be due to inherent
sensor limitations or visual teture.
- blur
- Edges in images are not "sharp" but are defocused by either optical imperfections
or by smoothing the image before edge detection.
- binary image
- An image in which each pixel is either a 0 or a 1.
- thresholding
- Convering a grey scale imageto a binary image based on whether a pixel in
the grey scale image has intensity greater than or less than or equal to a given
constant (the threshold).
- histogram
- Observed frequency distribution of image grey levels.
Connected Component Analysis
- 4-neighbors
- A pixel's horizontally and vertically adjacent pixels are its 4-neighbors
- 8-neighbors
- The 4-neighbors plus the diagonally adjacent pixels.
- 4-adjacent
- Given two sets of pixels, A and B, a is 4-adjacent to B is a pixel in A is a
4 neighbor of a pixel in B. Similar definition for 8-adjacent.
- path
- Sequence of pixels in which consecutive pixels are adjacent. Comes in 4
and 8 varieties.
- foreground
- Set of all 1's in a binary image
- connected
- Within a set S, if p and q are in S, then p is connected to q is there is a path
from p to 1; 4 and 8 versions.
- connected components
- The equivalence classes of the is-connect-to relation.
- background
- The component of 0's of a binary image that is adjacent to the edge of the image.
For this purpose, we envision the image as being surrounded by extra rows and columns
of 0's.
- boundary
- Boundary of a component of 1's is the subset of that component adjacent to a 0.
- interior
- The set difference between a component and its boundary
Edges and local features
- gradient
- vector whose first compoenent is the direction in which the first
derivative is highest and whose second component is the magnitude of the
first derivative in that direction.
- convolution
- Pointwise multiplication of each kxk neighborhood of a digital
image by a kxk kernel of convolution weights
- separable function
- A function F(x,y) is separable if it can be written as f(x,y)=g(x)h(y)
- non-maxima supression
- Procedure for thinning the responses of an edge detection. Retain as
edge points only pixels whose gradient magnitude is greater than the
magnitudes of its neighbors in the gradient direction.
- simple point
- A 1-point in a binary image is a simple point if changing its value froma 1
to a 0 does not change the number of connected components in its 3x3
neighborhood. There are 4 and 8 simple points.
- endpoint
- A 1 pixel in a binary image is an end point if it has exactly one
neighbor whose value is 1. There are 4 end points and 8 end points.
- junction
- A junction is a 1 pixel in a thinned binary image that has more than
2 neighbors that are 1's.
- curvature
- Rate of change of slope with respect to arc length.
Correlation
- template
- An image model for an object; can be grey level or binary.
- pyramid
- Multiresolution image representation in which an image at level i is obtained
from the image at level i-1 by smoothing (ordinarily Gaussian) and sampling. The
base of the pyramid, level 0, is the original image.
- sum of squared differences
- An image matching measure in which one integrates the squared difference in
grey levels bstween the image and a template.
- Hough transform
- Parameter space clustering algorithm for template matching. Originally
developed for line detection, it can be simply extendedto template matching.
- Distance Transform
- Given a binary image, its distance transform is an image in which each pixel is
assigned its distance from the nearest 1.
- chamfer matching
- Convolution of a binary edge template with a chamfer image for matching.
- Hausdorff matching
- a minimax distance metric for comparing a binary template to a binary image.