Exam 1, EE5357, Summer 2011
1. For nine (NF = 9) rectangular window features gkm , let the corresponding
overlapping windows wkm(x,y) be equal-size continuous functions of x and y, as
shown below. We want to construct disjoint windows wkm(x,y), whose unio
Exam 1, EE5357, Summer 2012
1. Images in an LPR system have non-overlapping blocks f(m,n) which are M
rows by N columns in size, so 1 m M and 1 n N. Let H(i) be the histogram
of a block, where H(i) is the number of pixels with grey level (brightness) i, a
Exam 2, EE5357, Summer 2011
1. Given some Gaussian training data for a classification problem, a Bayes-Gaussian
classifier can be designed. A functional link net (FLN) discriminant di(x) can be
designed through regression.
(a) Express the type B1 Bayes di
Exam 2, EE5357, Summer 2012
1. The data file D for designing a nearest neighbor (NN) classifier has Nv patterns,
each belonging to one of Nc classes. Nv(i) denotes the number of patterns in D from
the ith class. For J-fold validation, we divide D into J d
Exam 3, EE5357, Summer 2011
1. Let xk denote the kth feature vector calculated from an image i(x,y), where k
varies from 1 to K. Assume that the feature set operator FSkcfw_ is completely
different for different values of k.
(a) Let Nk = dim (xk). How can
Exam 3, EE5357, Summer 2012
1. a(x,y) and (x,y) are two low spatial frequency images that we might fuse into a
new image, which should also have low spatial frequencies.
(a) Assuming that a(x,y) and (x,y) are fused into a complex Pseudospectrum (PS)
s(x,y