OPEN CV image.at(b,a) function error in while loop while running second time

Question

My code is using OPENCV for image processing and I want to randomly create a feature set of points which are white. The program results in Assertion failed error when running the inner WHILE LOOP SECOND TIME. I have commented and tested the program to find exactly where is the problem. And i have found the problem with image.at(b,a) part of the code. It fails to run the second time. Although it runs perfectly fine the first time.
If i run the code with constant value instead of Variables b,a in the inner while loop, or with (j,j) variables it runs. I have used fout to print some debug statements. I am also attaching the .txt file output to help you see what's happening.

START OF THE OUTPUT FROM TXT FILE

start of while loop

inside the while loop

outside the while loop

[255, 255, 255]404 1173 start of the while loop

inside the while loop

END OF THE OUTPUT FROM TXT FILE

Pretty strange.Please help me figure this out!! I am pasting my code:

int main() {
    int i = 0;
    int a;
    ofstream fout("test.txt");
    int b;
    int j = 0;
    cv::Mat img;
    cv::Scalar* scalar_low;
    cv::Scalar* scalar_up;
    cv::Mat newimg;
    cv::Mat img2;
    stringstream ss;
    string s;
    scalar_low = new cv::Scalar(0, 30, 60);
    scalar_up = new cv::Scalar(20, 150, 255);
    int imgwidth;
    int imgheight;
    srand(112);
    while (j <= 45)
    {
    s="imagedata\\test"+ to_string(j);    
    s+=".jpg";
    img=imread(s);    
    cvtColor(img, img2, CV_BGR2HSV, 0);
    inRange(img2, *scalar_low, *scalar_up, newimg);            
    s="imagedata\\result"+ to_string(j);
    s+=".jpg";
    imwrite(s, newimg);
    imgwidth = newimg.cols;
    imgheight = newimg.rows;
    i = 0;
    while (i <= 499)
    {
        a = 0;
        b = 0;
               fout << "start of while loop \n";

            while ((newimg.at<Vec3b>(b,a) == Vec3b(0,0,0)))    
            {
                fout << "inside the while loop \n"
                a = rand() % imgwidth;
                b = rand() % imgheight;

            }
            fout << "outside the while loop \n"; 
            fout << a;
            fout << " ";
            fout << b;
            fout << " ";
            i++;

        }

        j++;
        fout << "\n";

    }


}

Answer 1

I am trying to build a random feature set of points which are white from a filtered image. That requires me to randomly generate points and check if they are white or not

You don't need to generate random points in the correct image range, and check if a pixel is white or not in the original image.

You can:

• get all white pixels in mask, with findNonZero
• random shuffle these points, with std::random_shuffle
• keep first N shuffled points

Masked image:

Random sampled N white points in mask:

Code:

#include <opencv2\opencv.hpp>
#include <vector>
#include <algorithm>
using namespace cv;
using namespace std;

int main()
{
    Mat3b img = imread("path_to_image");

    Mat3b hsv;
    cvtColor(img, hsv, COLOR_BGR2HSV);

    Mat1b mask;
    inRange(hsv, Scalar(0, 30, 60), Scalar(20, 150, 255), mask);

    // Get all white points in mask
    vector<Point> pts;
    findNonZero(mask, pts);

    int N = 1000; // number of random points to keep
    N = min(N, int(pts.size())); // not more than the actual white pixels

    // Random shuffle of white points
    random_shuffle(pts.begin(), pts.end());

    // Random subset of N white points in mask
    vector<Point> my_random_points(pts.begin(), pts.begin() + N);

    // Create a black image
    Mat1b random_in_mask(mask.rows, mask.cols, uchar(0));

    // Set selected random points to white
    for (const Point& pt : my_random_points)
    {
        random_in_mask(pt) = uchar(255);
    }

    imshow("Mask", mask);
    imshow("Random in Mask", random_in_mask);

    waitKey();

    return 0;
}

Answer 2

EDIT: InRange leaves newimg with type uchar or CV_8U, so change the condition to:

newimg.at<uchar>(a,b) == (uchar) 0

will solve the problem. Updated my answer according to comments below.

According to the this, the row is the dimension 0, which is the width. In at<>(i,j), i is along the dimension 0 j is along the 1. Your image is probably rectangular, so you second pair of rand()-s result cause bad a,b-s.

EDIT: In your code you have

while ((newimg.at<Vec3b>(b,a) == Vec3b(0,0,0)))    
{
     fout << "inside the while loop \n"
     a = rand() % imgwidth;
     b = rand() % imgheight;
}

Change this to

while ((newimg.at<Vec3b>(b,a) == Vec3b(0,0,0)))    
{
     fout << "inside the while loop \n"
     a = rand() % imgheight;
     b = rand() % imgwidth;
}

Or

while ((newimg.at<Vec3b>(a,b) == Vec3b(0,0,0)))    
{
     fout << "inside the while loop \n"
     a = rand() % imgwidth;
     b = rand() % imgheight;
}

I am saying this should work, not that it would.

EDIT #2: Explanation

You have a picture which looks like this for example:

-----------
|         |
|         |
|         |
|         |
|         |
|         |
-----------

But a and be with the at<>(b,a) usage order creates a different rectangle, so:

------------------
|         |      |
|         |   2  |
|     1   |      |
|---------|-------
|         |
|         |
-----------

Your first pair of random numbers have been generated to the (1) position, while the second pair is not on the image. You did not use srand(time(NULL)); so your random numbers are the same at each run.