减少多边形的边数

时间:2019-06-17 23:25:53

标签: matlab image-processing convex-polygon

假设我们有一个简单图形的图像,并且我们知道它是一个多边形,略有扭曲。有图像处理方法可以近似图形对象的原始参数吗?

下面的矩阵是通过代码创建的,然后缩小尺寸以显示第五个感兴趣的区域:

EnumeratedMask=bwlabel(Zdata<-.06);

0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   5   5   5   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   5   5   5   5   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   5   5   5   5   5   5   5   0   0   0   0   0   0   0   0
0   0   0   5   5   5   5   5   5   5   5   5   0   0   0   0   0   0   0
0   0   0   5   5   5   5   5   5   5   5   5   5   5   0   0   0   0   0
0   0   0   5   5   5   5   5   5   5   5   5   5   5   5   5   0   0   0
0   0   0   5   5   5   5   5   5   5   5   5   5   5   5   5   5   0   0
0   0   0   5   5   5   5   5   5   5   5   5   5   5   5   0   0   0   0
0   0   0   5   5   5   5   5   5   5   5   5   0   0   0   0   0   0   0
0   0   0   5   5   5   5   5   5   5   5   0   0   0   0   0   0   0   0
0   0   0   0   5   5   5   5   5   5   5   0   0   0   0   0   0   0   0
0   0   0   0   5   5   5   5   5   5   5   0   0   0   0   0   0   0   0
0   0   0   5   5   5   5   5   5   5   0   0   0   0   0   0   0   0   0
0   0   0   5   5   5   5   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   5   5   5   5   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   5   5   5   5   5   0   0   0   0   0   0   0   0   0   0   0
0   0   0   5   5   0   0   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0

下一步,我需要ABC / ABCD坐标来获得由这些点进一步定义的线的z轮廓。

2 个答案:

答案 0 :(得分:2)

这是上面Cris Luengo评论中建议的Ramer–Douglas–Peucker算法的实现。

这是答案的第一个版本的完整版本,该版本使用edge查找对象的边界。正如Cris Luengo在评论中指出的那样,bwboundaries是二进制图像的更好选择。 bwboundaries返回排序后的点,从而大大简化了代码。

以下代码执行以下操作:

1)使用bwboundaries找到对象的边缘。它们已经排序。

2)使用Ramer–Douglas–Pecker算法简化点列表。

由于我需要一些视觉提示来进行调试,因此代码打开了一个图,该图显示了正在发生的事情。

请注意,代码远未经过正确测试。

img = [...
    0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 5 5 0 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 5 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 5 5 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
    0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
    0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 5 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0];

watch = true;

if watch
    f = figure;
    subplot(1,2,1);
    imagesc(img);
end


% first, find the edges

b = bwboundaries(img);

% note that the first and the last point are the same.
% However they are already sorted.
x = b{1}(:,1);
y = b{1}(:,2);

edges = zeros(size(img));
edges(sub2ind(size(e), x,y)) = 1;

if watch
    ax = subplot(1,2,2);

    img_h = imagesc(edges);
    hold on;

end 

title('Performing Douglas-Peucker algorithm');

% Omit the last point for the algorithm.

points = l_DouglasPeucker( [x(1:end-1), y(1:end-1)], 1, watch);

title('Final result');
plot([points(:,2); points(1,2)], [points(:,1); points(1,1)]);



function res = l_DouglasPeucker( points, ep, watch )

    if nargin < 3
        watch = false;
    end

    if watch
        subplot(1,2,2);
        hold on;
        hp = plot(points(:,2), points(:,1), 'ko-');
        hp2 = plot([points(1,2) points(end,2)], [points(1,1) points(end,1)], 'r-');
    end
    distances = zeros(size(points,1),1);
    for i = 1:size(points,1)
        distances(i) = l_distance_to_line(points(1,:), points(end,:), points(i,:));
    end

    idx = find(distances == max(distances),1);


    if watch
        hp4 = plot(points(idx,2), points(idx,1), 'mo', 'MarkerFaceColor', [1,1,1]);
        pause(.5);
        delete(hp);
        delete(hp2);
        delete(hp4);
    end

    if max(distances) > ep
        res = [l_DouglasPeucker(points(1:idx,:), ep, watch); l_DouglasPeucker(points(idx:end,:), ep, watch)];
    else
        res = [points(1,:); points(end,:)];
    end


end

function d = l_distance_to_line(p1,p2,p)
% computes the distance of p to the line through by p1 and p2
% There might be much better implementations of this...

% we need 3-dimensional data for this

p1 = [p1(1), p1(2), 0];
p2 = [p2(1), p2(2), 0];
p = [p(1,1) p(1,2) 0];

a = p2 - p1;
b = p - p1;

d = norm(cross(a,b)) ./ norm(a);
end

enter image description here

答案 1 :(得分:2)

在帕特里克(Patrick)发表他的教育性答案之前,我已经开始工作,而且我还面临着Ramer-Douglas-Peucker算法的一个“问题”:按照定义,它保留了第一点和最后一点。 convhullboundary函数都从某处开始而不是总是在角落开始,这会触发误报。第三步和第四步解决了这个问题-另一点更有可能是真正的困境。

算法:

  1. 检测凸包/外点。
  2. 应用紧密匹配的Ramer-Douglas-Peucker算法(RDP)
  3. 倒退循环
  4. 应用RDP算法以更宽松的拟合去除假正角

代码:

eps1=2;
eps2=4;
img=[...
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
0 0 0 5 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0
0 0 0 5 5 5 5 5 5 5 5 5 5 5 0 0 0 0 0
0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 5 0 0 0
0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 5 5 0 0
0 0 0 5 5 5 5 5 5 5 5 5 5 5 5 0 0 0 0
0 0 0 5 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0
0 0 0 5 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
0 0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0
0 0 0 5 5 5 5 5 5 5 0 0 0 0 0 0 0 0 0
0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 5 5 5 5 5 0 0 0 0 0 0 0 0 0 0 0
0 0 0 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0];

[X,Y]=find(matr==5); % // find image coordinates of points of interest
is=boundary(X,Y,1);  % // find points on the boundary

xy=[X(is),Y(is)];    % // boundary points (step 1)

% // apply RDP algorithm (step 2)
xy=RDP(xy,eps1);

% // rewind the loop (step 3)
xy=xy([2:end-1 1:2],:);

% // apply the DRP algorithm (step 4)
xy=RDP(xy,eps2);

function[hull]=RDP(hull,eps)
    sp=hull(1,:);
    ep=hull(end,:);
    ip=hull(2:end-1,:);
    % // calculate distances of inner points from the first-last line
    dst=PerpDist(sp,ep,ip);
    % // find the point furthest from the f-l line
    [mx,mi]=max(dst);
    if mx>eps % // furthest point does not fit in - split and apply RDP recursively
        lp=[sp;ip(1:mi,:)];
        if size(lp,1)>2 % // there are points left to assess
            lp=RDP(lp,eps);
        end
        rp=[ip(mi:end,:);ep];
        if size(rp,1)>2 % // there are points left to assess
            rp=RDP(rp,eps);
        end
        hull=[lp;rp(2:end,:)]; % // concatenate the branches
    else % // the furthest poit fits in the limit, drop all inner points
        hull=[sp;ep];
    end
end

function[D]=PerpDist(A,B,C)
    D=nan(size(C,1),1);
    if A==B % // edge is defined by one point, use euclidean distance between points
        for PDi=1:size(C,1)
            D(PDi)=norm(C(PDi,:)-A);
        end
    else % // edge is a line, use eucleidean distance from a line
        for PDi=1:size(C,1)
            D(PDi)=abs(A(1)*(B(2)-C(PDi,2))+B(1)*(C(PDi,2)-A(2))+C(PDi,1)*(A(2)-B(2)))/norm(B-A);
        end
    end
end

enter image description here

点:形状内的点。
红场:boundary函数返回的第一个点和最后一个点。
绿线:第一次RDP模拟的结果。
洋红色线:最终三角形(原始形状为三角形)。

编辑:

  1. 留下寻找两个最远点以分裂环路的想法,因为即使对于三角形,它也会触发误报。从任意点开始,使用RDP两次即可达到目的。

积分:

史蒂夫·爱丁(Steve Eddins)的max_feret diameter
克里斯·伦戈(Cris Luengo)发表Ramer-Douglas-Peucker algorithm评论