phaseCorrelate fails with larger 16bit images

[rlexmann]

System Information

OpenCV version: 4.10.0
Operating System / Platform: Windows 11 Pro
Compiler & compiler version: MSVC 19.29.30154

Detailed description

With larger (~4000x4000px) uint16 images converted to float32, phaseCorrelate sometimes fails because infs and NaNs appear in divSpectrums().

Steps to reproduce

#include <opencv2/imgproc.hpp>
int main(int argc, char ** argv) {
    using namespace cv;

    // load
    Mat im = imread("<path/to/opencv_extra>/testdata/shared/baboon.png", IMREAD_GRAYSCALE);
    assert(im.type() == CV_8UC1);

    // convert to 32F, scale values as if original image was 16U
    constexpr auto u8Max = std::numeric_limits<std::uint8_t>::max();
    constexpr auto u16Max = std::numeric_limits<std::uint16_t>::max();
    im.convertTo(im, CV_32FC1, double(u16Max) / double(u8Max));

    // enlarge and create ROIs
    const auto w = im.cols * 5;
    const auto h = im.rows * 5;
    const auto roiW = (w * 2) / 3; // 50% overlap
    Mat imLarge;
    resize(im, imLarge, { w, h });
    const auto roiLeft = imLarge(Rect(0, 0, roiW, h));
    const auto roiRight = imLarge(Rect(w - roiW, 0, roiW, h));

    // correlate
    double response = 0.0;
    Point2d phaseShift = phaseCorrelate(roiLeft, roiRight, cv::noArray(), &response);
    assert(std::isnormal(phaseShift.x) || 0.0 == phaseShift.x);
    assert(std::isnormal(phaseShift.y) || 0.0 == phaseShift.y);
    assert(std::isnormal(response) || 0.0 == response);

    return 0;
}

Test data: baboon.png

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)