CSCE4240 - Exam 2 Flashcards
histogram algorithm
step 1.
create a one-dimensional array h of size L with initial value of zero
histogram algorithm
step 1. create one-dimensional array h of size L with initial value of zero
step 2.
loop through all pixels in the image A
histogram algorithm
step 1. create one-dimensional array h of size L with initial value of zero
step 2. loop through all pixels in the image
step 3.
for a pixel value v, increase the value at h(v) by one
histogram algorithm
step 1. create one-dimensional array h of size L with initial value of zero
step 2. loop through all pixels in the image
step 3. for a pixel value v, increase the value of h(v) by one
step 4.
continue until all pixels in the image A are visited
what is an image histogram
acts as a graphical representation of the tonal distribution in a digital image
what is histogram stretching
maps the value of all pixels to a new value that spans the full gray scale range
what is histogram equalization
maximize the usage of the full brightness range. maximum contrast is achieved when image histogram is uniform distribution
median filter algorithm
sort all the pixels in an increasing order and take the middle value
what are sampling rates
rate at which amplitude values are digitized from the original waveform
what is sinusoidal basis
use smoothly-varying sinusoidal patters at different frequencies, angles for basis of images - hadamard basis doesn’t capture real image gradients
what is magnitude
how much of a certain frequency component is in an image
what is phase
where that certain frequency lies
butterworts lowpass filter
introduces unwanted artifacts into the result. uses smooth transition
JPEG encoding
step 1.
transform RGB to YUV and subsample color
JPEG encoding
step 1. transform RGB to YUV and subsample color
step 2.
perform discrete cosine transform on 8x8 image blocks
JPEG encoding
step 1. transform RGB to YUV and subsample color
step 2. perform discrete cosine transform on 8x8 image blocks
step 3.
perform quantization
JPEG encoding
step 1. transform RGB to YUV and subsample color
step 2. perform discrete cosine transform on 8x8 image blocks
step 3. perform quantization
step 4.
zig-zag ordering and run-length encoding
JPEG encoding
step 1. transform RGB to YUV and subsample color
step 2. perform discrete cosine transform on 8x8 image blocks
step 3. perform quantization
step 4. zig-zag ordering and run-length encoding
step 5.
entropy encoding
what is quantization
aims at reducing the total number of bits by dividing each entry in the frequency space block by an integer then round
wavelet-based image fusion
step 1.
decompose images using wavelet transform
wavelet-based image fusion
step 1. decompose images using wavelet transform
step 2.
combine coefficients
- combine approximation subbing and the average
- select the maximum among detail subtends and put in the composite
wavelet-based image fusion
step 1. decompose images using wavelet transform
step 2. combine coefficients
step 3.
perform inverse wavelet transform on the composite wavelet matrix
noise-aware image fusion algorithm
step 1.
decompose images with wavelet transform
noise-aware image fusion algorithm
step 1. decompose images with wavelet transform
step 2.
compute the subbing noise variance On
noise-aware image fusion algorithm
step 1. decompose images with wavelet transform
step 2. compute the subbing noise variance
step 3.
estimate the threshold (lamda) that separate SIC and NIC
noise-aware image fusion algorithm
step 1. decompose images with wavelet transform
step 2. compute the subbing noise variance
step 3. estimate the threshold (lamda) that separate SIC and NIC
step 4/5.
- combine SIC
5. combine NIC
noise-aware image fusion algorithm
step 1. decompose images with wavelet transform
step 2. compute the subbing noise variance
step 3. estimate the threshold (lamda) that separate SIC and NIC
step 4/5. combine SIC and combine NIC
step 6.
synthesize a fused image from the subbing composite
JPEG2000 algorithm
1.
image tilting
JPEG2000 algorithm
1. image tilting
2.
DC-level shifting
JPEG2000 algorithm
1. image tilting
2. DC-level shifting
3.
components tranformation
JPEG2000 algorithm 1. image tilting 2. DC-level shifting 3. components tranformation 4.
wavelet transform
JPEG2000 algorithm 1. image tilting 2. DC-level shifting 3. components tranformation 4. wavelet transform 5.
quantization
JPEG2000 algorithm 1. image tilting 2. DC-level shifting 3. components tranformation 4. wavelet transform 5. quantization 6.
coefficient coding
Image morphology edge detection
1.
dilate the original image
Image morphology edge detection
1. dilate the original image
2.
subtract the original image from the dilated one
looks for particular pattern within the image
hit-and-miss transform
delete any such point that has more than one foreground neighbor, as long as doing so does not locally disconnect the region
thinning
suppresses the bright details that are smaller than the specified SE
opening
suppresses the dark details
closing
Chain Code
1.
find the top-left pixel on the boundary; call P0
Chain Code
1. find the top-left pixel on the boundary; call P0
2.
traverse the four neighborhood of the current pixel in the counter-clockwise order
Chain Code
1. find the top-left pixel on the boundary; call P0
2.traverse the four neighborhood of the current pixel in the counter-clockwise order
3
stop when current boundary pixel Pk equals to P1 and Pk-1 equals to P0
mean shift
1.
start from an arbitrary point in the distribution
mean shift
1. start from an arbitrary point in the distribution
2.
region of interest is a circle centered at this point
mean shift
1. start from an arbitrary point in the distribution
2. region of interest is a circle centered at this point
3.
on each iteration, find the center of mass for the ROI
mean shift
1. start from an arbitrary point in the distribution
2. region of interest is a circle centered at this point
3. on each iteration, find the center of mass for the ROI
4.
move the circle to this center
mean shift
1. start from an arbitrary point in the distribution
2. region of interest is a circle centered at this point
3. on each iteration, find the center of mass for the ROI
4. move the circle to this center
5.
continue the iterations until it convergences
hough transform
1.
discretize parameter space into bins
hough transform
1. discretize parameter space into bins
for each feature point in the image, put a vote in every bin in the parameter space that could have generated this point
hough transform
1. discretize parameter space into bins
2. for each feature point in the image, put a vote in every bin in the parameter space that could have generated this point
3
find bins that have the most votes
selecting seed without a-priori knowledge
compute the histogram and choose the grey values with the highest peak
otsu thresholding method idea
find the threshold that minimizes the weighted within-class variance and maximizes the between class variance
otsu thresholding method assumption
the histogram is a binomial distribution and the objects colors are mostly homogeneous
