System Information

OpenCV python version: 4.8.0.76
Operating System / Platform: macOS
Python version: 3.10.6

Detailed description

I noticed here that cvGetOptimalNewCameraMatrix returns (0,0,width-1,height-1) when the distortion parameters are zero.

Looking more closely at the code, I see that the internal function icvGetRectangles finds the rectangles by sampling points in the (0,width-1)x(0,hieght-1) range, and then computes the output ROI sizes as (maxX-minX,maxY-minY), which of course misses one pixel in each dimension!

I think the code in icvGetRectangles should add 1 pixel to both dimensions, because the max value should be included.

Steps to reproduce

Test code:

import cv2
import numpy as np
# zero distortion
distortion_params = np.zeros(8)
width=1920
height=1080
ffactor=1.6
# intrinsics with a centered principal point
K1 = np.array([[width*ffactor, 0, (width-1)*0.5],
              [0,width*ffactor, (height-1)*0.5],
              [0,0,1]])
# intrinsics with a non-centered principal point
K2 = np.array([[width*ffactor, 0, (width-1)*0.5+0.05],
              [0,width*ffactor, (height-1)*0.5-0.04],
              [0,0,1]])
for K in [K1,K2]:
    newK, roi = cv2.getOptimalNewCameraMatrix(K, distortion_params, (width, height), 0)
    print(np.linalg.norm(K-newK))
    print(roi)

Output:

4.547473508864641e-13
(0, 0, 1919, 1079)
4.547473508864641e-13
(0, 0, 1919, 1079)

both rois should be (0, 0, 1920, 1080).

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)