general_script.m

%% Problem1script
imname = 'test1.jpg';
imageData = im2double(rgb2gray(imread(imname)));

ohist = hog(imageData);

for k=1:9
    figure
    imshow(ohist(:,:,k));
end

%% Problem2script
% load a training example image
Itrain = im2double(rgb2gray(imread('pedSignTrain.jpg')));
% Itrain = im2double(imread('pedSignTemplate.jpg'));
%Itrain = im2double(rgb2gray(imread('pedSign3Train.jpg')));


% Itrain = im2double(rgb2gray(imread('test0.jpg')));
% nclick = 1;
% figure(1); clf;
% imshow(Itrain);
% [x,y] = ginput(nclick); %get nclicks from the user
% 
% %compute 8x8 block in which the user clicked
% blockx = round(x/8);
% blocky = round(y/8); 
% 
% % COMPUTE the template from user click: 
% % the following code assumes the template is 128x128 pixels  
% %   (16x16 hog blocks) so you will want to eventually modify
% %   it to handle different sized templates.
% %
% 
% %visualize image patches that the user clicked on
% figure(2); clf;
% for i = 1:nclick
%   patch = Itrain(8*blocky(i)+(-63:64),8*blockx(i)+(-63:64));
%   figure(2); subplot(3,2,i); imshow(patch);
% end

% compute the hog features
template = hog(Itrain);

figure; imshow(Itrain)

figure; clf; 
for i=1:size(template,3)
subplot(3,3,i); imagesc(template(:,:,i))
end

%
% load a test image
%
Itest= im2double(rgb2gray(imread('test3.jpg')));


% find top 5 detections in Itest
ndet = 5;
[x,y,score] = detect(Itest,template,ndet);

%display top ndet detections
figure(3); clf; imshow(Itest);
for i = 1:ndet
  % draw a rectangle.  use color to encode confidence of detection
  %  top scoring are green, fading to red
  hold on; 
  h = rectangle('Position',[x(i)-64 y(i)-64 128 128],'EdgeColor',[(i/ndet) ((ndet-i)/ndet)  0],'LineWidth',3,'Curvature',[0.3 0.3]); 
  hold off;
end

%% Problem 3 Script

Itrain = im2double(rgb2gray(imread('test1.jpg')));

template = hog(Itrain);
Itest = im2double(rgb2gray(imread('test0.jpg')));
currentTest= Itest;
resizeFactor = 0.7;

properTest = size(Itrain) < size(currentTest)
% properTest = size(Itrain) > size(currentTest);

blockWidth = 128;
level = 0;
xVals = []; yVals = []; scoreVals = [];
levelVals = [];
while(properTest(1) && properTest(2))
    
    % find top 5 detections in Itest
    ndet = 5;
    [x,y,score] = detect(currentTest,template,ndet);
    newX = x./(resizeFactor^level);
    newY = y./(resizeFactor^level);
    xVals = [xVals newX];
    yVals = [yVals newY];
    scoreVals = [scoreVals score];
    levelVal = (x.*0) + level;
    levelVals = [levelVals levelVal];

    %display top ndet detections
    %figure(3); clf; 
    figure
    imshow(currentTest);
    for i = 1:ndet
      % draw a rectangle.  use color to encode confidence of detection
      %  top scoring are green, fading to red
      hold on; 
      h = rectangle('Position',[x(i)-(blockWidth/2) y(i)-(blockWidth/2) blockWidth blockWidth],...
          'EdgeColor',[(i/ndet) ((ndet-i)/ndet)  0],...
          'LineWidth',3,'Curvature',[0.3 0.3]); 
      hold off;
    end
    
    blockWidth = blockWidth*resizeFactor;
    currentTest = imresize(currentTest,resizeFactor);
    properTest = size(Itrain)<size(currentTest);
    level = level + 1;
end

[val ind] = sort(score,'descend');
xFinal = xVals(ind);
yFinal = yVals(ind);
levelFinal = levelVals(ind);

%display top ndet detections
figure
imshow(Itest);
blockWidth = 128;
for i = 1:ndet
    
    blockWidth = 128*(resizeFactor^levelFinal(i));
    
  % draw a rectangle.  use color to encode confidence of detection
  %  top scoring are green, fading to red
  hold on; 
  h = rectangle('Position',[xFinal(i)-(blockWidth/2) yFinal(i)-(blockWidth/2) blockWidth blockWidth],...
      'EdgeColor',[(i/ndet) ((ndet-i)/ndet)  0],...
      'LineWidth',3,'Curvature',[0.3 0.3]); 
  hold off;
end

%% Problem 4 Script Picking Images
% this is the script used when you want
%       to pick the images. The script used
%       for the final result is the next one.

% load a training example image
Itrain = im2double(rgb2gray(imread('test3.jpg')));

%have the user click on some training examples.  
% If there is more than 1 example in the training image (e.g. faces), 
%   you could set nclicks higher here and average together
figure(1); clf;
imshow(Itrain);
numRects = 8;
numPosRects = 3; %number of initial rectangles which will be used for positive template
patches = cell(1,numRects);
widthValues = zeros(1,numRects);
heightValues = zeros(1,numRects);

for num = 1:numRects
    rect = getrect(figure(1));
    xmin = floor(rect(1));
    ymin = floor(rect(2));
    width = floor(rect(3));
    height = floor(rect(4));
    patch = Itrain(ymin:(ymin+height),xmin:(xmin+width));
    widthValues(num) = width;
    heightValues(num) = height;
    patches{num} = patch;
end

aspectRatioValues = widthValues./heightValues;
resizeHeight = mean(heightValues);
avgAspectRatio = mean(aspectRatioValues);
resizeWidth = resizeHeight*avgAspectRatio;
numHeightBlocks = floor(resizeHeight/8);
numWidthBlocks = floor(resizeWidth/8);
resizeHeight = numHeightBlocks*8;
resizeWidth = numWidthBlocks*8;

resizedPatches = zeros(resizeHeight,resizeWidth,numRects);

for num = 1:numRects
   figure
   resizedPatch = imresize(patches{num},[resizeHeight resizeWidth]);
   resizedPatches(:,:,num) = resizedPatch;
   imshow(resizedPatch)
end

% compute the hog features
%f = hog(Itrain);

negTemplate = zeros(numHeightBlocks,numWidthBlocks,9);
posTemplate = zeros(numHeightBlocks,numWidthBlocks,9);
for i = 1:numPosRects
   f = hog(resizedPatches(:,:,i)); 
   posTemplate = posTemplate + f;
end
for i = numPosRects+1:numRects
   f = hog(resizedPatches(:,:,i)); 
   negTemplate = negTemplate + f;
end
posTemplate = posTemplate/numPosRects;
negTemplate = negTemplate/(numRects-numPosRects);
template = posTemplate-negTemplate;

%
% load a test image
%
Itest= im2double(rgb2gray(imread('test4.jpg')));

% find top 5 detections in Itest
ndet = 5;
[x,y,score] = detect(Itest,template,ndet);

%display top ndet detections
figure(3); clf; imshow(Itest);
for i = 1:ndet
  % draw a rectangle.  use color to encode confidence of detection
  %  top scoring are green, fading to red
  hold on; 
  h = rectangle('Position',[x(i)-64 y(i)-64 128 128],'EdgeColor',[(i/ndet) ((ndet-i)/ndet)  0],'LineWidth',3,'Curvature',[0.3 0.3]); 
  hold off;
end

%% Script for getting negative images
% This gets random image patches from the test images, which were
%   later checked to make sure they don't contain the pedestrian sign
%   Used to gather images for problem 4

Itrain = im2double(rgb2gray(imread('test0.jpg')));
imsize = size(Itrain);
height = imsize(1)-168;
width = imsize(2)-168;

numPics = 100;
randXvals = floor(rand(1,numPics)*width) + 84;
randYvals = floor(rand(1,numPics)*height) + 84;
for picNum = 1:numPics
   imname = strcat('prob4negTrain/negPatchX_',num2str(picNum),'.jpg'); 
   patch = Itrain(randYvals(picNum)-84:randYvals(picNum)+84,...
       randXvals(picNum)-84:randXvals(picNum)+84);
   imwrite(patch,imname,'JPEG');
end



%% Problem 4 Script Actual Run

% load a training example image

resizeHeight = 168;
resizeWidth = 168;
numHeightBlocks = floor(resizeHeight/8);
numWidthBlocks = floor(resizeWidth/8);
numPosRects = 5;

posPatches = zeros(resizeHeight,resizeWidth,numPosRects);

posPatches(:,:,1) = im2double(rgb2gray(imread('prob4posTrain/pedSign2Train.jpg')));
posPatches(:,:,2) = im2double(rgb2gray(imread('prob4posTrain/test3train.jpg')));
posPatches(:,:,3) = im2double(rgb2gray(imread('prob4posTrain/test4train.jpg')));
posPatches(:,:,4) = im2double(rgb2gray(imread('prob4posTrain/test5train.jpg')));
posPatches(:,:,5) = im2double(rgb2gray(imread('prob4posTrain/test6train.jpg')));

posTemplate = zeros(numHeightBlocks,numWidthBlocks,9);
for i = 1:numPosRects
   f = hog(posPatches(:,:,i)); 
   posTemplate = posTemplate + f;
end
posTemplate = posTemplate/numPosRects;

negTemplate = zeros(numHeightBlocks,numWidthBlocks,9);

%comment out this block to get one with just 5 positive templates
files = dir('prob4negTrain');
filesSize = size(files);
numFiles = filesSize(1);
images = zeros(168,168,100);
for index = 3:numFiles
   imname = strcat('prob4negTrain/',files(index).name);
   image = im2double(imread(imname));
   imageToUse = image(1:168,1:168);
   f = hog(imageToUse);
   negTemplate = negTemplate+f;
end
negTemplate = negTemplate/(numFiles-2);

template = posTemplate-negTemplate;

%
% load a test image
%
Itest= im2double(rgb2gray(imread('test6.jpg')));

% find top 5 detections in Itest
ndet = 5;
[x,y,score] = detect(Itest,template,ndet);

%display top ndet detections
figure(3); clf; imshow(Itest);
for i = 1:ndet
  % draw a rectangle.  use color to encode confidence of detection
  %  top scoring are green, fading to red
  hold on; 
  h = rectangle('Position',[x(i)-64 y(i)-64 128 128],'EdgeColor',[(i/ndet) ((ndet-i)/ndet)  0],'LineWidth',3,'Curvature',[0.3 0.3]); 
  hold off;
end