Description
(1) (5 points) For this problem you will do the second additional task associated with Practical
Lesson 7.6.2. Specifically, using the nearest neighbor interpolation example that we went over in
class (in the book-related file: NNInterpolation_7.m), which applied a rotation image transformation with Nearest Neighbor interpolation, adapt it to instead use a bilinear interpolation. Plot
the nearest neighbor result along with the bilinear interpolation result, side-by-side.
(2) (5 points) This problem is similar to part of the first additional exercise for Practical Lesson
9.9.2. You may not use any built in MATLAB functions for computing histograms, entropy or
mutual information. Note, we went through the code for this practical lesson in class. Compute
and display entropy for each image (the CT and T1), which was defined in class as
H =
X
i
pi
log
1
pi
.
Be sure to leave out any probabilities of 0 in the computation, since this will give you a Nan or inf
result. Use the natural log, which is simply log in MATLAB. Next, for angles between -90 to 100
(in increments of 10, as we did in class) compute the joint entropy, which was defined in class as
H(CT, T1) = X
i,j
pi,j log
1
pi,j
,
were CT and T1 refer to the CT and T1 images used and pij are the probabilities in the joint
histogram. Be sure you use the probabilities in your calculation. Change the figure heading of each
joint histogram to show the joint entropy value as well as the normalized mutual information,
H(CT) + H(T1)
H(CT, T1) .
You may need to use the newline command (\n) and %.4g in your sprintf command to make
the figure headings readable.
1
(3) (5 points) Use the test.txt and train.txt data that were distributed with the homework assignment (read them in using load(’path/to/file’, ’-ascii’)). These data consist of normalized
handwritten digits scanned from envelopes by the U.S. Postal Service where each image is a 16 ×16
grayscale image. The files contain 257 columns, where the first column indicates the written digit
and the following 256 columns represent the grayscale values of the written digit. If you read
test.txt with the variable name test, then test(1,1) shows the digit represented in the image
and imagesc(reshape(test(1,2:end), 16,16)’) will plot the first image, which you will see is
an image of the digit that matches test(1,1). Using all 256 features, write the code for K-nearest
neighbors to perform this 10 category classification. You must write this code yourself and cannot
use any KNN algorithms either built into Matlab or supplied online. Since there are more than
2 categories, you cannot directly use the code from lecture, but must adapt it to use Euclidean
distance for the identification of neighbors and classification is determined by majority vote, randomly selecting in the case of ties. Using 5 nearest neighbors, classify the test data according to
the training data. Create a bar plot that shows the classification accuracy for each digit.
2