RGB Normalizied give Black image - c++

Question

Here , I am trying to do normalize RGB image.
Here is my code.
cv::Mat channels[3], normalize_rgb;
split(image, channels);
for (int i = 0; i < image.size().height; i++)
{
for (int j = 0; j < image.size().width; j++)
{
int b = (int)(image).at<cv::Vec3b>(i, j)[0];
int g = (int)(image).at<cv::Vec3b>(i, j)[1];
int r = (int)(image).at<cv::Vec3b>(i, j)[2];
double sum = b + g + r;
double bnorm = b / sum * 255;
double gnorm = g / sum * 255;
double rnorm = r / sum * 255;
channels[0].at<uchar>(i, j) = bnorm;
channels[1].at<uchar>(i, j) = gnorm;
channels[2].at<uchar>(i, j) = rnorm;
}
}
merge(channels, 3, normalize_rgb);
normalize = normalize_rgb.clone();
Problem: after normalizing r,g, b value is giving me very small value which turns to 0. and hence I get black image.
Anyone can please help me to figure out problems. thanks

Answer 1

Change image to double first. It is black because when you divide int by int you get zero. And then you're trying to set uchar = double which is completely wrong.
Try:
double bnorm = double (b) / sum * 255;

Answer 2

You need to divide the sum by the number of pixels.
In a 1000x1000 image you want to divide by the average, not by one million times the average.
Note that dividing by the average may give you values bigger than 255 (values higher than the average) so clamping is required after the scaling. Probably a better solution would be aiming at the average becoming 128, not 255 with
double bnorm = b * k;
double gnorm = g * k;
double rnorm = r * k;
where k is computed once per image as
double average = sum / (image.size().height*image.size().width);
double k = 127.0 / average;
but note that this can still produce values higher than 255 in output.
If you want to stretch you need to normalize with the maximum value, not with the average and in this case clamping is not needed.

Answer 3

I found a link which related to RGB Normalization, there is a mathematical list in the article.
So actually, u need to divide a var total, and each total is just the sum of each pixel, not the whole image(width() * height()). Maybe you should try this:
for (int i = 0; i < image.size().height; i++)
{
for (int j = 0; j < image.size().width; j++)
{
int b = (int)(image).at<cv::Vec3b>(i, j)[0];
int g = (int)(image).at<cv::Vec3b>(i, j)[1];
int r = (int)(image).at<cv::Vec3b>(i, j)[2];
int sum = b + g + r;
double bnorm = b / sum * 255;
double gnorm = g / sum * 255;
double rnorm = r / sum * 255;
channels[0].at<uchar>(i, j) = bnorm;
channels[1].at<uchar>(i, j) = gnorm;
channels[2].at<uchar>(i, j) = rnorm;
}
}