What do the values of the mask parameter returned by findHomography represent?

Question

I am using the function findHomography of OpenCV with the RANSAC method in order to find the homography that relates two images linked with a set of keypoints.

The main issue is that I haven't been able to find anywhere yet what are the values of the mask matrix that the function outputs.

The only information that I know is that 0 values are outliers, and non zero values are inliers. But what does it mean the inliers value? Does anyone know?

Piece of code where I call findHomography:

cv::Mat H12;
cv::Mat mask;

H12 = cv::findHomography(FvPointsIm1, FvPointsIm2, mask, CV_RANSAC, 5); 
ui->Debug_Label->setText(Mat2QString(mask));

Cyclamen

Are you asking about the meaning of the inlier/outlier distinction, or about the numerical values in the mask? – United

About the numerical values in the mask. – Cyclamen

Do you ever get values other than 0 or 1? – United

Outliers are 0. Inliers get values, most of them very small. They order of magnitude is between e^-200 and e^-400. – Cyclamen

Could you post the code where you call findHomography and then examine the mask values? It would be nice to know the type of your mask input. – United

I've just updated it in the top. Thank for the interest. – Cyclamen

Answer 1

The mask returned by findHomography is an 8-bit, single-channel cv::Mat (or std::vector<uchar>, if you prefer) containing either 0 or 1 indicating the outlier status.

EDIT: You access each element of the mask by calling .at<double>, which is leading to the confusing output. You should be using .at<uchar>, which will interpret the matrix value correctly.

United

To convert the values of the mask to QString I use the following for each matrix field: ` QString::number(mask.at<double>(i,j)) ` – Cyclamen

Nice, that was the mistake. Thank you very much! – Cyclamen

I cannot print the value correctly by using: mask.at<uchar>(0, 0). It prints a small matrix icon with values "0 0, 0 1"... Do you know why? – Droplet

Okay... It is just that I call findHomography() exactly as he does, and when I print cout << mask.at<uchar>(0, 0) << endl; as you proposed, I don't get a 0 or a 1. I will see to ask another question if you think it is different as here. – Droplet

@user1368342 I've reproduced your problem. Cast the result to an int to see the numerical value, like so: cout << (int)mask.at<uchar>(0,0) << endl; – United

@Aurelius, you are right: #24457288 Here is the question and also the answer. It was quite trivial after all. :P – Rural

Answer 2

I used the findHomography method after applying keypoint matching.

• Inliers are matched keypoints that are calculated to be true positives (correct matches);
• Outliers are matched keypoints that are calculated to be false positives (false matches).

Then you can use the mask output to extract the subset of correct matches from all matches.

• There is an example in Python 3.6 & OpenCV 3.4.1:

  good_kp = [gray_kp[m.queryIdx].pt for m in good_matches]
  correct_matched_kp = [good_kp[i] for i in range(len(good_kp)) if mask[i]]
  

Alonzo