Speed optimization of function ICAngles in orb.cpp

[spdfghi]

System information (version)

• OpenCV => 4.2
• Operating System / Platform => Windows 64 Bit
• Compiler => Visual Studio 2017

Detailed description

This issue is expalantion of pr #25337 closed for creating new one.

Function ICAngles calculates the centroid of a pixel in the surrounding circular area. Since the process area is symmetric, consider four points: $v_1=a[-x_0,-y_0], \quad v_2=a[x_0,-y_0], \quad v_3=a[-x_0,y_0], \quad v_4=a[x_0,y_0]$, then opencv now use a for loop iterating x, and process pixel with same absolute y value together:

 for (int v = 1; v <= half_k; ++v)
{
    // Proceed over the two lines
    int v_sum = 0;
    int d = u_max[v];
    for (int u = -d; u <= d; ++u)
    {
        int val_plus = center[u + v*step], val_minus = center[u - v*step];
        v_sum += (val_plus - val_minus);
        m_10 += u * (val_plus + val_minus);
    }
    m_01 += v * v_sum;
}

which means sum for point 1,3 is $sum_{1,3}=y_0(v_3-v_1)-x_0(v_3+v_1)$, and for point 2,4 is $sum_{2,4}=y_0(v_4-v_2)+x_0(v_4+v_2)$. Now I want to combine sum for these four points as $sum_{1,3}+sum_{2,4}=y_0(v_3-v_1+v_4-v_2)+x_0(v_4+v_2-v_3-v_1)$

total operation nums for these four points:

	mult	add
no optimization	8	8
opencv now	4	8
pr 25337	2	8

Reduced ratio of operation nums is 1/6, consistent with time test in pr 25337.

Is there any problems? If not, I will make a new pr again.

Steps to reproduce

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues,
  forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files: videos, images, onnx, etc