Question

I want to get the maximum values from two images, while preserving other information.

For example, I have two images in HSL format, and I want the resulting image to consist of the brightest pixels. (this is an example, in reality I can’t convert to RGB, finding max, and converting back)

So I cannot use cvMax because it will separately give me the maximum H, maximum S and maximum L.

What I need is an image with the maximum L from either image, accompanied by the H and S values (or other information) from the same image.

For example, consider these 2×1 pixel images, with their HSV scalars:

IplImage* img1 = cvCreateImage(cvSize(2, 1), IPL_DEPTH_32F, 3);
cvSet2D(img1, 0, 0, cvScalar(1, 2, 45)); // Remember: HSV values
cvSet2D(img1, 0, 1, cvScalar(3, 4, 123));
IplImage* img2 = cvCreateImage(cvSize(2, 1), IPL_DEPTH_32F, 3);
cvSet2D(img1, 0, 0, cvScalar(5, 6, 114));
cvSet2D(img1, 0, 1, cvScalar(7, 8, 33));

IplImage* result = cvCreateImage(cvSize(2, 1), IPL_DEPTH_32F, 3);

//Do something to get the brightest pixels

CvScalar px0 = cvGet2D(result, 0, 0);
CvScalar px1 = cvGet2D(result, 0, 1);

cout << px0.val[0] << " " << px0.val[1] << " " << px0.val[2] << endl;
cout << px1.val[0] << " " << px1.val[1] << " " << px1.val[2] << endl;

Output should become:

5 6 114
3 4 123

Is there a neat ‘OpenCV’ way of doing this, instead of making my own, possibly slower, algorithm that parses each pixel?

---

Just an idea, but can I do something like this?

• Split them into 6 grayscale images, (actually, I already have this situation, so I don’t need to split)
• Take both V images,
• Generate a per-element map (1U or 8U image) where a ‘0’ or ‘1’ value means the highest value was in the first or second image respectively.
• Use this to combine selected pixels from the other planes.

The bold step is where I’m stuck.

Answer 1

There is no build-in opencv function to do it, but it’s not thah difficult to write your own.

You can simply

for(i=0;i<h;i++)
   for (j=0;j<w;j++)
      dst[j+w*i] = (src1[j+w*i]>src2[j+w*i])?src1[]:src2[];

Of course, some little optimization can be done, like precalculating j+w*i, or using pointer access instead of array indexes.