How to recognize the stars in the photo of the night sky using OpenCV?

Question

I want to write a program that will recognize stars in the photo of the night sky and mark them. I'm new in DSP and would like to ask how to implement this. I sketched a draft code:

#include "main.hpp"

using namespace std;
//using namespace cv;

int main(int argc, char *argv[])
{
    const char *imageName = (argc >= 2) ?   argv[1] : "4.jpg";
    int64_t processTime;
    cv::Mat imageSource, result1, result2, grayScaledImage;

    /// Create and initialize the kerenel matrix
    cv::Mat kernelMatrix = (cv::Mat_<char>(3, 3) << 0, -1, 0,
                                                -1, 5, -1,
                                                0, -1, 0);

    cout << "Start..." << endl;

    if ((argc == 3) && !(strcmp("G", argv[2])))     imageSource = cv::imread(cv::samples::findFile(imageName), cv::IMREAD_GRAYSCALE);
    else                                            imageSource = cv::imread(cv::samples::findFile(imageName), cv::IMREAD_COLOR);

    /// Check for errors while open
    if (imageSource.empty())
    {
        cerr << "Can't open image [" << imageName << "]" << endl;
        exit(EXIT_FAILURE);
    }

    processTime = cv::getTickCount();                                           /// Start timer
    //img_proc::sharpening(imageSource, result1);                                 /// Process image
    cv::cvtColor(imageSource, grayScaledImage, cv::COLOR_BGR2GRAY);
    cv::filter2D(grayScaledImage, result1, imageSource.depth(), kernelMatrix);      /// Process image
    processTime = (cv::getTickCount() - processTime) / cv::getTickFrequency();  /// Stop timer and fix the process time

    /// Create windows for pictures
    cv::namedWindow("Source image", cv::WINDOW_AUTOSIZE);
    cv::namedWindow("Result image (1)", cv::WINDOW_AUTOSIZE);
    /// Show images
    cv::imshow("Source image", imageSource);
    cv::imshow("Result image (1)", result1);
    cv::waitKey(0);

    /// The second method
    processTime = cv::getTickCount();                                           /// Start timer
    cv::Sobel(result1, result2, CV_32F, 1, 0);                                  /// S
    processTime = (cv::getTickCount() - processTime) / cv::getTickFrequency();  /// Stop timer and fix the process time

    cv::namedWindow("Result image (2)", cv::WINDOW_AUTOSIZE);
    cv::imshow("Result image (2)", result2);
    cv::waitKey();

    double minVal, maxVal;
    cv::Mat newMat;
    cv::minMaxLoc(result2, &minVal, &maxVal);
    result2.convertTo(newMat, CV_8U, (255.0 / (maxVal - minVal)), (-minVal * 255.0 / (maxVal - minVal)));

    cv::namedWindow("newmat", cv::WINDOW_AUTOSIZE);
    cv::imshow("newmat", newMat);
    cv::waitKey();


    return 0;
}

As a result I get the following images (on the left - the initial image, on the right - the final image): result images

But I want to highlight the stars, for example, with a yellow circle on a grayscale image. How I can do it?

Why did you apply these two filters specifically? What did you hope to get from them? I would try a simple local maximum search on the input image. — Nico Schertler, Nov 18 '19 at 01:43
Nico Schertler, I used filter2D function to sharpen image, and then I tried to use something that mark the stars. But in my opinion, the next two filters (sobel, minMaxloc) not needed for this task. — Ruslan, Nov 18 '19 at 01:58
Did you try a [local maximum search](https://stackoverflow.com/questions/5550290/find-local-maxima-in-grayscale-image-using-opencv)? — Nico Schertler, Nov 18 '19 at 02:02
There's also method called `H-maxima`. Local maxi can give you some false positives, but when it comes to stars, it wouldn't pain so much. There're nice docs on [skimage](https://scikit-image.org/docs/dev/auto_examples/segmentation/plot_extrema.html) site. Additionally, if you're new to image processing, I'd rather advise you go with `python` + `scikit-image`. `opencv` docs are rather disappointing. — Piotr Rarus, Nov 18 '19 at 15:51

score 2 · Accepted Answer · answered Nov 18 '19 at 20:51

Here is a MATLAB solution:

I = imread('Stars.jpg'); %Read image from file
I = I(:, 1:floor(size(I,2)/2-40), :); %Crop the relevant part (left side).
J = rgb2gray(I); %Convert RGB to Grayscale.
BW = imbinarize(I); %Convert to binary image (you have the option to manually select the threshold).
BW2 = bwmorph(BW2, 'shrink', Inf); %Shrink clusters until single pixel is left.

%The shrink operation wasn't good enough, fill hols, and shrink again.
BW3 = bwfill(BW2, 'holes');
BW4 = bwmorph(BW3, 'shrink', Inf);

%Find coordinates where pixel value is 1
[Y, X] =  find(BW4);

%Mark each coordinate with yellow circle.
K = insertMarker(I, [X Y], 'o', 'size', 5, 'color', 'yellow');
figure;imshow(K); %Display result.

Sorry for using MATLAB, but with OpenCV it's going to take me 100 hours...

Result:

MATLAB has a profound documentation, so you can learn about each operation by Google searching.

How to recognize the stars in the photo of the night sky using OpenCV?

1 Answers1