Draw polygons using libvips

[ido-nuc]

Hi!

I am currently working on a pathology-related project where I am dealing with (very) large-sized images. I chose libvips (specifically the pyvips binding) due its capability to handle large images efficiently. My pipeline includes rendering shapely geometries (i.e., points and polygons) & blending multiple rendered images together.

However, after searching through the libvips documentation, I noticed that there doesn't seem to be a draw_polygon operation. This leads me to two questions:

1. Why libvips does not have a draw_polygon operation? If it does exist and I missed it, please let me know.
2. Given the current set of operations, what would you suggest as the best way to implement the pipeline I discussed above? Any advice on how to approach this and any alternative methods for rendering shapely geometries and blending images together would be greatly appreciated.

Thank you for your assistance!

Best,
Ido

[jcupitt]

Hi @ido-nuc,

The simplest solution is just to draw them with SVG:

import pyvips

x = pyvips.Image.svgload_buffer(b"""
    <svg viewBox="0 0 200 200">
      <circle r="100" cx="100" cy="100" fill="#900"/>
    </svg>
""")

x.write_to_file("x.png")

To make:

The SVG loader is progressive, so it can draw SVGs of any size (up to 32k x 32k pixels) using only a modest amount of memory. You can make masks by just taking the alpha, you can combine masks with the boolean operators, and you can render stacks of RGBA images with composite.

The 32k pixel limit is unfortunate :( It comes from cairo, sadly, and isn't really fixable. We are considering switching to another SVG renderer, but no one's done the work yet.

Don't use the draw_circle etc. operators, They do not work well with large images.

Another option is to use xyz. This makes an n-dimensional index image where each pixel contains its own coordinates. You can then use the usual arithmetic, relational and boolean operators to draw any shape. It's surprisingly quick, and scales well to images to any size. Coming up with the right maths to draw the shape you want is annoying, however.

x = pyvips.Image.xyz(501, 501) - [250, 250]
x = (x[0] ** 2 + x[1] ** 2) ** 0.5 < 200

Will draw a circle with a radius of 200 in the centre of a 500 x 500 pixel image, for example.

[lejsekt]

In case somebody needs a similar operation in C#:

using var x = Image.Xyz(501, 501) - new[] { 250, 250 };
using var distance = x[0] * x[0] + x[1] * x[1];
using var pow = distance.Pow(0.5);
using var result = pow <= 200;