1 简介
Harris 角点检测算法是由 C.Harris 和 J.Stephens 在 1988 年提出,此算法的原理是建立在 Moravec 算法的基础上进行大幅改进的,相比之前的算法更具明显的优势。Moravec如上文中的介绍说明,相比 Harris 检测算法而言,
Moravec 检测算法仅仅只是是在水平、垂直、对角线和反对角线四个方向上对灰度变化的信息进行处理,而 Harris 检测算法受到数学理论泰勒(Taylor
)级数的启发,以此展开思路,已不只是限定于四个方向,而是计算窗口内沿任意方向的灰度变化情况,最终利用数学解析式进一步确定所需特征点。并且算法引入高斯平滑算子,更加优化了算法自身的鲁棒性。不仅如此,Harris
检测算法还受到信号处理中自相关函数的启迪,从而使用了自相关函数相关联的 M 矩阵。
该算法中 M 矩阵的特征值即自相关函数的一阶求导函数,判断该点是否为特征点即看此处像素点的曲率值是否够高。
正在上传…重新上传取消正在上传…重新上传取消正在上传…重新上传取消
正在上传…重新上传取消正在上传…重新上传取消
2 部分代码
function [imgout]=warpTheImage(H,img1,img2)
tform = maketform('projective',H');
img21 = imtransform(img2,tform); % reproject img2
%
[M1 N1 dim] = size(img1);
[M2 N2 dimk1] = size(img2);
% do the mosaic
pt = zeros(3,4);
pt(:,1) = H*[1;1;1];
pt(:,2) = H*[N2;1;1];
pt(:,3) = H*[N2;M2;1];
pt(:,4) = H*[1;M2;1];
x2 = pt(1,:)./pt(3,:);
y2 = pt(2,:)./pt(3,:);
up = round(min(y2));
Yoffset = 0;
if up <= 0
Yoffset = -up+1;
up = 1;
end
left = round(min(x2));
Xoffset = 0;
if left<=0
Xoffset = -left+1;
left = 1;
end
[M3 N3 dimk2] = size(img21);
rowBegin=max(up,Yoffset+1); %overlap Area
columnBegin=max(left,Xoffset+1);
rowEnd=min(up+M3-1,Yoffset+M1);
columnEnd=min(left+N3-1,Xoffset+N1);
imgout(up:up+M3-1,left:left+N3-1,:) = img21;
overlapAreaP2=imgout(rowBegin:rowEnd,columnBegin:columnEnd,:);%pixel values of overlap area from P2
% img1 is above img21
imgout(Yoffset+1:Yoffset+M1,Xoffset+1:Xoffset+N1,:) = img1;
overlapAreaP1=imgout(rowBegin:rowEnd,columnBegin:columnEnd,:);
overlapArea=imgout(rowBegin:rowEnd,columnBegin:columnEnd);
[overRowLength,overColumnLength]=size(overlapArea);%overlap Row and Column length
distFromBound1OneLine=(overColumnLength-1:-1:0);%this is just one line
distFromBound1=repmat(distFromBound1OneLine,overRowLength,1); %Replicate and tile it to the size of the overlapArea. Because the same column has the same distance to the boundary
distFromBound2OneLine=(0:overColumnLength-1);
distFromBound2=repmat(distFromBound2OneLine,overRowLength,1);%this the dist from boundary 2
% blending
% blendingImg(:,:,:)=(overlapAreaP2(:,:,:).*distFromBound2+overlapAreaP1(:,:,:).*distFromBound1)/(overColumnLength-1);
% imshow(blending)
overlapAreaP2=double(overlapAreaP2);
overlapAreaP1=double(overlapAreaP1);
blendingImg=zeros(overRowLength,overColumnLength,3);
for i=1:3
blendingImg(:,:,i)=(overlapAreaP2(:,:,i).*distFromBound2+overlapAreaP1(:,:,i).*distFromBound1)/(overColumnLength-1);
end
blendingImg=uint8(blendingImg);
% imshow(blendingImg);title('after blending');
imgout(rowBegin:rowEnd,columnBegin:columnEnd,:)=blendingImg;
end
3 仿真结果
4 参考文献
[1]贾莹. 基于Harris角点检测算法的图像拼接技术研究[D]. 吉林大学.
博主简介:擅长智能优化算法、神经网络预测、信号处理、元胞自动机、图像处理、路径规划、无人机等多种领域的Matlab仿真,相关matlab代码问题可私信交流。
部分理论引用网络文献,若有侵权联系博主删除。