Class6-Image Segmentation

1.Graylevel Thresholding

作用：就是在灰度图中将灰度值小于阈值的像素保留，否则置为0.
实现代码：

function[]=Graylevelthresholding()

clear, clc, close all

% Load test image
img = imread('peter.png');

% Threshold
level = 105;
bwImg = img < level;
holeImg = img .* uint8(bwImg);

% Show images
subplot(1, 3, 1), imshow(img); title('Original Image');
subplot(1, 3, 2), imshow(bwImg); title('Thresholded Image');
subplot(1, 3, 3), imshow(holeImg); title('Binary Map \times Original');

% Save images
imwrite(bwImg, 'Graylevel_Thresholding_thresholded.png');
imwrite(holeImg, 'Graylevel_Thresholding_blend.png');

在ppt中，提了一个问题就是如何将图二的黑洞填上？
感觉是下面介绍的方法三

2.最大类间方差法

顾名思义该方法利用目标区域与背景区域之间的方差最大的思想，达到分割图像的目的。也就是说，选取最佳阈值T时，目标与背景之间的方差值最大，小于阈值T的区域为D1,大于阈值的区域为D2，如此一来即可将需要的区域区分开来。
$${\sigma }_{\text{within}}^2(T)=\frac{N_{Fgmd}(T)}{N}{\sigma }_{Fgnd}^2(T)+\frac{N_{Bgmd}(T)}{N}{\sigma }_{Bgrnd}^2(T)$$

于是，Ostu算法就是找到一个阈值T，使得${\sigma }_{\text{betwan}}^2(T)=\frac{N_{\text{Fgad}}(T)\cdot N_{Bgnd}(T)}{N^2}{\left({\mu }_{Fgnd}(T)-{\mu }_{Bgnd}(T)\right)}^2$最大
幸运的是，Matlab提供了这个函数的！
课堂代码如下：

% Load test image
% img = imread('peter.png');
% img = imread('brain.jpg');
% img = rgb2gray(imread('news.png'));
function []=Global_Thresholding()
img = rgb2gray(imread('front.png'));
% img = imread('paper.png');

% Perform Otsu thresholding
level = graythresh(img); % chooses Otsu threshold
otsuThresh = round(level * 255);
bwImg = im2bw(img, level);

% Show images
subplot(1, 3, 1), imshow(img); title('Original Image');
subplot(1, 3, 2), imshow(bwImg); title('Globally Thresholded Image');
subplot(1, 3, 3), imshow((1-bwImg) .* im2double(img)); title('Overlay');
axes('Parent', figure, 'FontSize', 18);
[counts,x] = imhist(img);
bar(x, counts); hold on;
h = plot(otsuThresh*ones(1,100), linspace(0,max(counts)), 'r-');
% title('Graylevel Histogram');
axis([0 255 0 max(counts)]);
set(gca, 'FontSize', 18);
set(gcf, 'Color', 'white');

% Save images
imwrite(bwImg, 'Global_Thresholding_bw.png');
saveas(gcf, 'Global_Thresholding_hist.png')

其中，需要解释一下im2bw是将像素大于阈值的全部置为255，小于的全部置为0
实验效果：

3.局部适应阈值

Non-uniform areas: apply Otsu’s method (based on local histogram)
Uniform areas: classify the entire area as foreground or background based on mean value