Using vision.CascadeObjectDetector to find rotated faces and findSURFFeatures to align them with original - doesn't work!

1 次查看（过去 30 天）

mark palmer 2020-5-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/522331-using-vision-cascadeobjectdetector-to-find-rotated-faces-and-findsurffeatures-to-align-them-with-ori

编辑： mark palmer 2020-5-2

I'm trying to use a face detector to locate rotated faces in an image and calculate their bounding box locations in the original unrotated image. When I run this code, the rotated face does not line up with the original. Here is the image I am using, blanked out in part to make the test code only search for one face.

Not sure why this doesn't work...

clear all
% Create a face detector    
fD1 = vision.CascadeObjectDetector;
I = imread('visionteam+rotfaces1.jpg');
 
% This detects the first face and puts a bounding box around it.
bboxes = step(fD1, I);  % Detect face
fI = insertObjectAnnotation(I, 'rectangle', bboxes, 'a');
figure, imshow(fI)
title('Detected face'); 
% bboxes x,y,sizex, sizey = 189    41    41    41
 
% Now rotate the image and detect faces.
I2 = imrotate(I,-90);
bboxes2 = step(fD1, I2);  % Detect face
fI2 = insertObjectAnnotation(I2, 'rectangle', bboxes2, 'b');
figure, imshow(fI2)
title('Detected face'); 
% bboxes2 = 202   354    42    42
 
% Now use detectSURFFeatures to match up original face and rotated face
% DOESN'T WORK!
original = rgb2gray(I);
x12 = bboxes(1,2)+1:bboxes(1,2)+bboxes(1,4);
y12 = bboxes(1,1)+1:bboxes(1,1)+bboxes(1,3);
oROI = original(x12,y12);
distorted = imrotate(oROI,-90);
 
ptsOriginal  = detectSURFFeatures(original);
ptsDistorted = detectSURFFeatures(distorted);
[featuresOriginal,  validPtsOriginal]  = extractFeatures(original,  ptsOriginal);
[featuresDistorted, validPtsDistorted] = extractFeatures(distorted, ptsDistorted);
% Match features by using their descriptors.
indexPairs = matchFeatures(featuresOriginal, featuresDistorted);
 
% Retrieve locations of corresponding points for each image.
matchedOriginal  = validPtsOriginal(indexPairs(:,1));
matchedDistorted = validPtsDistorted(indexPairs(:,2));
 
% Show putative point matches.
figure;
showMatchedFeatures(original,distorted,matchedOriginal,matchedDistorted);
title('Putatively matched points (including outliers)');