Image segmentation index and Matlab code implementation

Reprinted from: https://blog.csdn.net/m_buddy/article/details/78807478

1. Segmentation Metrics

After completing the image segmentation task, it is necessary to evaluate the performance of the segmentation results. Generally, the indicators used are IOU and false positive rate . For example, look at the segmentation control image 
 
below. In the third image above, the red one is the segmentation result, the green one is the marked image, and the yellow one is the overlapping area. Then the specific judgment is this 
 
IOU: 
 
false positive rate: 

2. Matlab implementation code

% 函数功能：计算两张图像的IOU
% lable_img：输入的图像A，只包含0和255像素值的图像
% res_img:输入的图像B，也同样只包含0和255像素值的图像
% 返回值：iou两张图像的iou；yl假阳性率
function [iou,yl] = Calc_IOU(lable_img, res_img)
[rows, cols] = size(lable_img);

% 计算总面积
%total_area = rows * cols;

% 统计lable_img、res_img中255像素值的个数
lable_area = 0; % 标记出来的面积
res_area = 0;   % 分割出来结果的面积
intersection_area = 0; % 相交区域的面积
combine_area = 0;      % 两个区域联合的面积

% 开始计算各部分的面积
for i = 1: 1: rows
    for j = 1: 1: cols
        if lable_img(i, j)==255 && res_img(i, j)==255
            intersection_area = intersection_area + 1;
            lable_area = lable_area + 1;
            res_area = res_area + 1;
        elseif lable_img(i, j)==255 && res_img(i, j)~=255
            lable_area = lable_area + 1;
        elseif lable_img(i, j)~=255 && res_img(i, j)==255
            res_area = res_area + 1;
        end
    end
end
combine_area = combine_area + lable_area + res_area - intersection_area;

% 得到IOU
iou = double(intersection_area) / double(combine_area);
fprintf('IOU: %f\n', iou);
% 得到假阳性率
yl = double(res_area - intersection_area) / double(combine_area);
fprintf('假阳性率为：%f\n', yl);

end