Description
1) Histogram Equalization
Use imread to read in the image lungs.jpeg.
This is a grayscale image but is stored as an RGB image with 3 color planes, so you can use:
a = lungs(:,:,1);
imshow(a)
ghe = histeq(a);
imshow(ghe)
to define a grayscale image ”a” and compute the global histogram equalized version (ghe) of it.
a) Read what the function ”cumsum” does. Carry out the following sequence of steps:
[n,y] = imhist(a,256);
map = cumsum(n);
map = map / max(map);
plot(y,map)
imshow(a, [map map map])
Explain what ”map” represents, what the plot shows, and what image is being shown with this
imshow command.
b) The dark background on this image is hurting the histogram equalization of this image.
Manipulate the histogram in some way which will help this problem, and explain what you
did. Then use cumsum to generate a new map, and show your image with this new map.
Plot the new map on the same plot as the previous one, and interpret the differences in your
equalized images relative to the differences in these plots. Include your plot with 2 curves and
your explanation of results.
2) Noise Cleaning
Load the files cleanbaby.mat (the baby image without added noise), babyS.mat (the baby image
with salt noise added, in the form of little vertical streaks), and baby2.mat (the baby image with
both salt noise and a low level of Gaussian noise).
a) Use medfilt2 with a 1×3 median filter to clean up the babyS image, and also to clean up the
baby2 image. Likewise use a 3×3 median filter to clean them up. In all four cases, use the
function immse to compute the mean squared error (MSE) between your filtered output and
the cleanbaby image.
You should find that the 3×3 filter does better (in terms of MSE) than
the 1×3 filter for one of the noisy images but the 1×3 is better for the other. Explain why this
occurs, and provide your matlab commands and your MSE values.
b) Try the sequential use of a median filter and a spatial averaging filter on the baby2 image that
has both noise types. Provide your Matlab code, the MSE between your output image and the
cleanbaby image, and a discussion of your results.
3) Unsharp Masking
a) You are to spatially enhance an image using unsharp masking. Consider the following MATLAB function:
function im_out = unsharp( im_in, maskA, weight )
[a,b] = size( maskA );
maskB = zeros( size( maskA)); maskB(ceil(a/2),ceil(b/2)) = 1;
maskC = maskB – maskA;
maskD = maskB + weight * maskC;
im_out = conv2(im_in,maskD,’valid’);
Here im in is the input image and im out is the output image. Suppose maskA is a small
odd-sized lowpass filter mask, and weight is a positive number. What kind of masks are masks
B, C and D? Using the discussion from class on separating an image into lowpass and highpass
components, explain how this function performs edge sharpening.
b) In unsharp masking you can choose which lowpass filter to use and how much weight to give
the highpass part. We will investigate the effect of these on the resulting image. First, create
a test image of size 128 × 128 that consists of a ramp and simple step function, as follows:
tst=ones(128,1)*[64*ones(1,32) (64:4:188) 192*ones(1,32) 64*ones(1,32)];
This has four equal sized areas (from left to right): first, 32 columns with value 64, then
32 columns of ramp going from 64 to 192, then 32 columns with value 192 and finally 32
columns with value 64.
Try a few combinations of low pass filters and weights. Vary the
size of the mask (e.g., 3 × 3, 5 × 5, maybe even 7 × 7) and the entries in the mask (e.g.,
unweighted averaging vs. strongly center-weighted averaging) and vary the extra weight given
to the highpass part. Discuss the results in each case, and note any trends you see that arise
from varying the parameters. Also look at a slice of the filtered images:
plot(tst(64,:));
Include in your homework a few of the plots of horizontal slices through the filtered images,
with discussions of the trends.
c) For a real-world example at larger size, read in the image ”blurry-moon.tif” and use highboost
filtering on it starting with the following Gaussian lowpass filter:
f = 1/159
2 4 5 4 2
4 9 12 9 4
5 12 15 12 5
4 9 12 9 4
2 4 5 4 2
Try different boost values. Provide your code and your output image, and comment on your
result.
4) Order of Operations (this is not a Matlab problem)
An image has been corrupted with a low level of additive Gaussian noise, and with 1% of salt
and pepper noise. The noise is independent from pixel to pixel. We will process the image with
3 operations in some order: contrast manipulation, median filtering, and spatial averaging filtering.
Consider that the image is stored as values ranging from 0 to 1.
• The contrast manipulation (CS) consists of remapping an input value s to an output value r
according to r =
√
s
• The median filter (MF) is a 3×3 median filter
• The spatial averaging (SA) filter is the unweighted 3×3 mean filter
The 3 operations can be done in 6 possible orders:
A: MF → SA → CS
B: CS → MF → SA
C: SA → CS → MF
D: MF → CS → SA
E: SA → MF → CS
F: CS → SA → MF
a) Which of these systems, if any, will produce an output image that is exactly the same as the
output from another system? Explain why?
b) Which of these systems, if any, do you expect to do better at noise removal? You are not
being asked for a complete ranking of the systems, but rather just to point out (and explain)
if certain placement orders clearly outperform some other ones for this noise removal task.