% playing with max/min filtering
% playing with convolution and masks

addpath /research/src/DNAdesign
addpath /research/src/AFMutils


% rizal's beautiful image
%
[im,in] = showafm('~winfree/Research/DNA/AFM_old/data.rizal/02-07-20/QBOX.097',0); 
im = flipud(im);
figure(1); 
    imagesc(im);
    truesize
    colormap gray
    title 'raw image'
    
% scale so that all elements lie in [0,1]
low = min(min(im));
high = max(max(im));
im = im - low;
im = im / high;

% % goal: to learn to use min and max filtering
% % improve image quality
% %
% % slow version maximum filtering
% im_max = nlfilter(im, [5,5], 'max(x(:))');
% figure(2)
%     imagesc(im_max)
%     truesize
%     colormap gray
%     title 'maxima filter'
%     
% % slow version minimum filtering
% im_min = nlfilter(im, [3,3], 'min(x(:))');
% figure(3)
%     imagesc(im_min)
%     truesize
%     title 'minima filter'
%
% These are unreasonably slow.
% 
% % min/max filtering is too responsive to noise?
% h = fspecial('average', [3,3]);
% im_ = imfilter(im, h);
% figure(2)
%     imagesc(im_ave)
%     truesize
%     colormap gray
%     title 'averaging filter'
% % blah
% 
% % would it help to apply a gaussian filter first?
% h = fspecial('gaussian', [5,5],1);
% im_ = imfilter(im, h);
% figure(2)
%     imagesc(im_ave)
%     truesize
%     colormap gray
%     title 'gaussian filter'
% % blah
% 
% % faster version
% im_min = colfilt(im_, [3,3], 'sliding', @min);
% im_max = colfilt(im_min, [3,3], 'sliding', @max);
% im_new = im_max;
% figure(3)
%     imagesc(im_new)
%     truesize
%     colormap gray
%     title 'min max 3,3'
% % these perform infinitely better than nlfilter :)
% 
% % will thresholding help at all?
% min_ = min(min(im_new));
% max_ = max(max(im_new));
% threshold = min_ + (max_-min_)*.75
% figure(4)
%     imagesc(im_new>threshold)
%     truesize
%     colormap gray
%     title 'threshold image'
% % not really, image is too segmented into blocks. tiles are cut in half by
% % thresholding. 
% 
% % trying to emphasize maximums and deemphasize minimums
% im_min = colfilt(im_, [5,5], 'sliding', @min);
% move = max(max(im_min));
% im_min = im_min - move;
% 
% im_max = colfilt(im_min, [5,5], 'sliding', @max);
% move = min(min(im_max));
% im_max = im_max - move;
% 
% im_new = im_min + im_max;
% figure (2)
%     imagesc(im_new)
%     truesize
%     colormap gray
%     title 'worth trying'
% 
% % thresholding again
% up = max(max(im_new));
% down = min(min(im_new));
% range = up - down;
% threshold = down + .7 * range;
% figure (3)
%     imagesc(im_new > threshold )
%     truesize
%     colormap gray
%     title 'threshold it'
% % looks pretty nice to me...
% 
% let's try a convolution on this image now
% miserable failure.
%     
% convolution appears to work better than this max and min filtering does
% in locating and emphasizing image features /tiles.

% % something completely unrelated that i wanted to play with.
% selection = roipoly(im);
% figure (2)
%     imagesc(selection);
%     truesize
%     colormap gray
%     title 'selected region'
%     % add feedback loop, user accepts selection or wants to restart?
%     
% % make a mask using the user's selected area.
% % mask = selection .* im;
% mask = im;
% % mask constrained to recentagle.
% [r,c] = size(mask);
% % if a column is empty, remove it
% for i = c:-1:1
%     temp = sum( selection(:,i) );        % will == 0 if empty
%     if (temp == 0)
%         mask(:,i) = [];
%         selection(:,i) = [];
%     end
% end
% 
% % if a row is empty, remove it.
% for i = r:-1:1
%     temp = sum( selection(i,:) );
%     if (temp == 0)
%         mask(i,:) = [];
%         selection(i,:) = [];
%     end
% end

% % do not assume mask is symetric in any way.
% % filter mask -- too noisy!
% h=fspecial('gaussian');
% mask = imfilter(mask,h,'same');
% figure (2)
%     imagesc(mask);
%     truesize
%     colormap gray
%     title 'selected mask'  
% % mask is still too messy, cannot get good results. 

% % make up a fake mask, see how it works. 
% mask = ones([20,10]);
% mask(1,:) = 0;
% mask(20,:) = 0;
% mask(:,1) = 0;
% mask(:,10) = 0;
% % works better than random mask.... blech.

mask = [zeros(20,6),  ones(20,6)]

imc = conv2(im,mask,'same');
figure(3);
    imagesc(imc);  
    colormap gray; 
    truesize
    title 'mask applied to raw image' 
    

pause;
close all;