Computer Vision cs682                       Due: Mar. 17, 2009
Spring 2009
Dr. Duric

                        Homework #2 (35p)

Your second homework involves extending your vision system for locating
humans in a digital image. This homework has three stages:
1. (15p) Computing edges for a background image and for an image with a
   human in it.
2. (10p) Subtracting edge images to create a difference image and
   classifying edges in the difference image. 
3. (10p) Connected component analysis to identify the human in the image.

1. Computing edges for a background image and images with humans

   You should use the programs provided with this description to gain
   experience and some intuition with edge detection. You should apply
   these programs to all supplied images. Those programs are
   BiSmooth.m (binomial smoothing), Gradient.m (gradient computation),
   ColorGradients.m (color gradients), NMS.m (non-maxima suppression), 
   and Hysteresis.m (hysteresis thresholding as in Canny edge detector). 
   These programs work for gray-level images only. To compute edges in
   color images you should use ColorGradients function. You should read the 
   programs to become familiar with some simple Matlab functions used in
   them. In addition you should use Matlab function quiver to display
   image gradients. Here is an example:
    > A = imread('fr05.tif');
    > A1 = BiSmooth(A,8);
    > [Gx,Gy] = Gradient(A1);
    > m1=NMS2(Gx,Gy,3);
    > m1 = Hysteresis(m1,3);
    > [I,J] = find(m1>0);  % row and column indexes
    > In = find(m1>0);     % indexes in a vector form
    > [M,N] = size(m1);
    > quiver(J,M-I+1,Gx(In),Gy(In),1)
   Create a web page and post the original and gradient images (vector
   fields) - email the URL to me at zduric@cs.gmu.edu. You can save
   your figures using print command in tiff, jpeg, or eps formats. You
   should post the results for at least three different images.

2. Subtracting edge images to create a difference image and
   classifying edges in the difference image

   In this part you will create difference images for Dx (x component
   of gradient) and Dy (y component of gradient). These difference
   images will have high values when gradients in the background and 
   the current images are different and low values when they are
   similar. After computing difference images you will classify edges
   into three types:
    Background edges: these are edges that look similar in the
       background and current images.
    Occluded edges: these are background edges that are not visible in
       the current image; possibly because they are occluded by a
       foreground object
    Occluding edges: these are foreground edges that do not exist in
       the background image

   How would you do this classification?

3. (10p) Connected component analysis to identify the human in the
   image

   You will compute connected components for all your classified
   edges. What edges correspond to a human in an image?
   Note that you can box the edges corresponding to a human in the
   image? Are they connected? 
   Could you suggest a background subtraction algorithm that would use 
   edge subtraction?

Image Files 

Image files can be obtained from 
      http://www.cs.gmu.edu/~zduric/cs682/Images/

Program Files

Matlab files can be obtained from 
      http://www.cs.gmu.edu/~zduric/cs682/MatlabFiles/

Submitting your homework

You will post on your web page the following:
i. description of all algorithms used in your program,
ii. complete testing results including labeled images of 
humans and any intermediate images that describe your work,
iii. discussion of possible applications for your program. Email the 
url to zduric@cs.gmu.edu