% CS143, Intro to Computer Vision, Michael Black

% Matlab code from the Sept 17 class
% Goal: illustrate image smoothing with a Gaussian

% read in and display the peppers image
cd c:\black\Teaching\cs143.03\data\
im=double(imread('peppers256.tif'));
figure(1); imagesc(im);
colormap gray
axis off
axis image

% add random noise to the image
nim=im+20*(rand(size(im))-0.5);
figure(2); imagesc(nim);
colormap gray
axis off
axis image

% look at the difference between the original and
% noisy images.  Note that there is no structure to
% the noise (it is independent and identically distributed).
figure(3);
colormap gray;
imagesc(nim-im);

% Construct a really big Gaussian filter with large variance.
figure(4);
G=fspecial('gaussian',15,2.5)
surf(G)

% Construct a 3X3 Gaussian 
G=fspecial('gaussian',5,1)
surf(G)


% convolve an image with the filter we created
cim=conv2(nim,G,'same');
imagesc(cim)

cim2 = imfilter(nim,G,'symmetric','conv');

% look at the difference between the convolved image
% and the original
% Notice that in addition to reducing the noise,
% it is no longer unstructured.  By applying this
% isotropic filter, we smoothed the edges and these
% now appear in the "noise".
imagesc(im-cim2)
colormap gray

% look at the difference from the noisy image. 
imagesc(nim-cim)