Image Processing How to Apply gradient [-1 | 0 | 1 ] to RGB image - c++

Question

I need to apply gradient operator to RGB bitmap image. It works for 8 bit image but having the difficulty in implementing same for 24 bit image. Here is my code. Can anyone see how
to correct the zorizontal gradient operation to RGB image.
if (iBitPerPixel == 24) ////RGB 24 bits image
{
for(int i=0; i<iHeight; i++)
for(int j=1; j<iWidth-4; j++)
{
//pImg_Gradient[i*Wp+j] = pImg[i*Wp+j+1] - pImg[i*Wp+j-1] ;
int level = pImg[i*Wp+j*3+1] - pImg[i*Wp+j*3-1] ;
pImg_Gradient[i*Wp+j*3] = level;
// pImg_Gradient[i*Wp+j*3] = level;
// pImg_Gradient[i*Wp+j*3+1] = level;
// pImg_Gradient[i*Wp+j*3+2]= level;
}
for(int i=0; i<iHeight; i++)
for(int j=0; j<iWidth; j++)
{
// Copy the convetred values to original image.
pImg[i*Wp+j] = (BYTE) pImg_Gradient[i*Wp+j];
}
//delete pImg_Gradient;
}

Answer 1

Unfortunately, it is not clear how to define a gradient of an RGB image. The best way to go is to transform the image into a color space that separates intensity from color, such as HSV, and compute the gradient of the intensity component. Alternatively, you can compute the gradient of each color channel separately, and then combine the results in some way, such as taking the average.
Also see Edge detectors for RGB images?

Answer 2

In order to calculate the Gradient of an image (Which is a vector) you need to calculate both the horizontal and vertical derivative of the image.
Since we're dealing with a discrete image we should use Finitie Difference approximations of the derivative.
There are many ways to approximate, many of them are listed on the Wikipedia Pages:
http://en.wikipedia.org/wiki/Finite_difference
http://en.wikipedia.org/wiki/Finite_difference_method
http://en.wikipedia.org/wiki/Finite_difference_coefficients
Basically those are Spatial Coefficients hence you can define a filter using them and just filter the image.
This would be the most efficient way to calculate the gradient.
So, all you need is to find a library (Such as Open CV) which supports filtering images and you're done.
For color images, usually, you just calculate the Gradient per Color Channel.
Good Luck.

Answer 3

From your code; you are trying to calculate gradient from RGB but there is nothing to indicate how RGB is stored in your image. A complete guess is that in your image you have BGRBGRBGR...etc.
In that case your code is getting the gradient from the green channel, then storing it in the red of the gradient image. You don't show the gradient image being cleared to 0 - if you don't do this then it will probably be full of junk.
My suggestion is to convert to a greyscale image first; then you can use your original code.
Or calculate a gradient for each colour channel.