System Information

OpenCV version: 4.2.0
Operating System / Platform: Ubuntu 20.04
Compiler & compiler version: GCC 9.3.0

Detailed description

icvGetRectangles estimates the inscribed (inner) and circumscribed (outer) rectangles for an undistorted image by generating a uniform grid of pixel coordinates and undistorting each coordinate. However, the grid includes a row and column of pixels outside of the image. Here it considers the maximum horizontal and vertical pixel coordinates to be imgSize.width and imgSize.height, respectively, but they should be imgSize.width-1 and imgSize.height-1.

This method is used in getOptimalNewCameraMatrix and stereoRectify so fixing it will slightly improve undistortion and rectification.

I discovered this when using getOptimalNewCameraMatrix: passing an ideal camera matrix with zero distortion returns a slightly different camera matrix (see below).

Steps to reproduce

    // Simple camera matrix with no distortion
    const cv::Size img_size = {101, 101};
    const cv::Matx33d K = {1, 0, 50, 0, 1, 50, 0, 0, 1};
    const auto dist_coeffs = cv::Matx14d::zeros();

    const bool center_principal_point = false;
    const double alpha = 0;
    cv::Rect roi;

    // K_new should be the same as K
    const cv::Matx33d K_new = cv::getOptimalNewCameraMatrix(K, dist_coeffs, img_size, alpha, img_size, &roi, center_principal_point)

The new 'optimal' camera matrix should be the same as the original but it is not:

Optimal new K:
[0.9900990128517151, 0, 49.50495064258575;
 0, 0.9900990128517151, 49.50495064258575;
 0, 0, 1]
ROI:
[100 x 100 from (0, 0)]

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)