Name: _________________________________
Score: /19
CSE 5524 Computer Vision for HCI
Homework Assignment #3
Due: See Carmen for due date
1) Generate a 3-level Gaussian pyramid (original image is level-0) and the corresponding
Laplacian pyramid of an image (select one from the web, make it grayscale). Use the
formula in the notes to first determine a viable image size (use N=3, and pick NC and
NR), and crop the image (if needed) to test the pyramid code. Use a=0.4 for the
Gaussian mask – use separable masks! Write/use functions for properly reducing and
expanding an image. Write your own interpolation function - do not use
Matlab/Python in-built interpolation functions (e.g., interp2). Lastly, perform a
reconstruction of the original (cropped) image using the Laplacian pyramid. [8 pts]
2) Using the grayscale images (walk.bmp, bg000.bmp) provided on the WWW site,
perform background subtraction 1 (abs diff) to extract the object. Make sure your
image is of type double! Experiment with thresholds and discuss. [2 pts]
3) Using the grayscale images (walk.bmp, bg[000-029].bmp) provided on the WWW
site, perform background subtraction 2 using statistical distances. Experiment with
thresholds and discuss. [5 pts]
4) Dilate your best binary image resulting from problem 3) using: [1 pt]
d_bsIm = bwmorph(bsIm, ‘dilate’);
5) Next perform a connected components algorithm, and keep only the largest region in
L (save/display as an image). [1 pt]
[L, num] = bwlabel(d_bsIm, 8);
Turn in a report containing all code, printouts of images, and discussion of results. Make a script
to do all tasks and call needed functions. Upload your report, code, and selected images to
Carmen. [2 pts]
MATLAB help: Loading several ordered grayscale images:
for i=1:N
filename = sprintf('myimage_%03d.bmp', i-1); % if starts with 000
Im(:,:,i) = double(imread(filename));
end
The %03d gives a zero-padded number (e.g., 001), and the image storage is now a 3-D cube. To get image 3 from the cube, use myIm = Im(:,:,3).
With such a cube, MATLAB operations such as ‘mean’ can be told to work along certain dimensions of the cube.