Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Material Type: Paper; Professor: Jones; Class: Electrical Machine Design; Subject: Electrical and Computer Engr; University: University of Illinois - Urbana-Champaign; Term: Unknown 2000;
Typology: Papers
1 / 30
This page cannot be seen from the preview
Don't miss anything!
ECE598FL Readings on
Computer Vision
IJCV
, 2000.
Comparison of Graph Cuts with Believe Propagationfor Stereo, using Identical MRF Parameters; byTappen, Freeman,
ICCV
, 2003.
ECE598 Readings onComputer Vision
ECE598 Readings onComputer Vision
^
^
ECE598 Readings onComputer Vision
(b) and (c): Interpolation by pixel replication(d) and (e): Cubic-spline interpolation(f) and (g): After sharpening the interpolated
image in (d) and (e).
(h) and (i): Training-based Super-
Resolution Algorithm.
ECE598 Readings onComputer Vision
^
Ck
^
th^
( )
(^
)
K k
C k^
k x g
Z
P^
1 1 x
ECE598 Readings onComputer Vision
Slide taken from: Bayesian Learning of Directed and Undirected Graphical Models; by ZoubinGhahramani. 2003.
( )
(^
)
K k
C k^
k x g
Z
P^
1 1 x
ECE598 Readings onComputer Vision
No theoreticalsupport, butexperimentalresults are good.
TreeNetwork
Some theoreticaljustifications.
MarkovNetworkwithoutloops
Useful in nets withloops?
MAP Estimate for nets without loops
[^
]^
(^
)^
(^
)
(^
)∏
∏
≠ k
k j
j j
x
j
jl
l j
j j
k j
x k j
y x
x
M y x x x M
j k
max arg
max MAP
(^
)^
(^
)
(^
) (
∏ )∏
≠^ k
k j
j j j
x
j
jl
l j
k k j k
x k j
M x y P x P x M x y P x x P M
j k
max arg
ˆ
max MAP
ECE598 Readings onComputer Vision
^
^
ECE598 Readings onComputer Vision
ECE598 Readings onComputer Vision
ECE598 Readings onComputer Vision
Highest resolution frequency band (
H
) is
conditionally independent of the lowerfrequency bands (
L ), given the mid-frequency
band (
M
):
Statistical relationships between image bandsare independent of image contrast, apart froma multiplicative scaling.
(^
)^
(^
) M H P L M H P |
, |^
=
ECE598 Readings onComputer Vision
ECE598 Readings onComputer Vision
^
^
(^
)^
(^
)
(^
) (
∏ )∏
≠^ k
k j
j j j
x
j
jl
l j
k k j k
x k j
M x y P x P x M x y P x x P M
j k
max arg
ˆ
max MAP
ECE598 Readings onComputer Vision
^
Training set: 80 images(animals and urban) a.^
Nearest neighborsolution b.^
st 1 Iteration c.^
nd 2 Iteration d.^
rd 3 Iteration e.^
Input data f.^
Desired output
