Evrything circle related is manual instead using openCV

[atsju]

So today we have this fillcircle method

DFTFringe/circleoutline.cpp

Line 31 in c20c76e

void fillCircle(cv::Mat &m, double cx, double cy, double rad, void *color) {

George is doing round masks #272
And in both case we do manual circles and radius.

The status is we cannot use [cv::circle](https://docs.opencv.org/4.x/d6/d6e/group__imgproc__draw.html#gaf10604b069374903dbd0f0488cb43670) because we use double coordinates for center and radius.

What I just discovered: the cv::circle method has

shift : Number of fractional bits in the coordinates of the center and in the radius value.

@dale can you confirm today coordinates and radius are only one half of a pixel?
This would mean we could use int coordinates with shift of 1. And not use double for this.

I don't want to change every coordinate system but I suspect we can use more opencv functions thanks to this and it will be easier to maintain + faster in some cases + bug free because do do not reinvent the wheel.

[gr5]

I can confirm that centerx, centery, radius could be integers if you doubled them.

That function you link to, cv::circle, doesn't fill the circle.

Dale uses inside/outside circle concept in many places. Instead of having that code in many places it should be a member function of the wavefront class (or a generic parent class) that determines if a point is masked. The function could use a cached mask or it could calculate it on the fly from outline and region information. Regions could be selected or ignored. Anyway this is just ideas for if I rewrote DFTF from scratch, lol. Also I'd implement unlimited shapes such as rectangles, not just circles and ellipses and annular circles.

[atsju]

CV::circle can fill circles. But not outside of circles. But you can easily make a disk mask and fill an other matrix from this mask. Outside or inside.

I will look where it makes sense to replace. We don't want to rewrite coordinate system in int.

[gr5]

I don't think replacing the way we make masks will speed anything up much.

When I was doing the gaussians, a complete matrix copy (with my own personal for loop etc) of 360,000 doubles (600X600) was too fast to worry about compared to the roughly 100 million multiplies to do the gaussian blur.

If we are going to speed something up it would probably be in the code that makes the zernike shapes. Or caching or something. But making a mask is relatively instant.