#image-processing #pixel #image #processing #imaging #painting

picture

a fast and flexible image manipulation crate

2 unstable releases

0.2.0 Mar 16, 2023
0.1.0 Nov 4, 2022

#346 in Images

Apache-2.0

110KB
2.5K SLoC

picture

a fast and flexible image manipulation crate.

warning: wip

this crate is still a WIP. while it's foundations are there (but not stable), it's still missing many convenience and quality of life features - mostly formats related.

it's also in necessity of tests™ (there's very few).

todo

  • better & expanded image formats support (currently very half-baked and only supports png and qoi)
  • add way more tests
  • try to reduce usage of unsafe blocks

fractal example

based on the fractal example of the image crate:

use picture::{formats::png::PngEncoder, prelude::*};
use std::io::Write;

fn main() {
    let (width, height) = (1024, 1024);
    let mut img = Rgb8Img::new(width, height);

    let scalex = 3.0 / width as f32;
    let scaley = 3.0 / height as f32;

    for ((x, y), pixel) in img.pixels_with_coords_mut() {
        let cx = y as f32 * scalex - 1.5;
        let cy = x as f32 * scaley - 1.5;

        let c = num_complex::Complex::new(-0.4, 0.6);
        let mut z = num_complex::Complex::new(cx, cy);

        let mut g = 0;
        while g < 255 && z.norm() <= 2.0 {
            z = z * z + c;
            g += 1;
        }

        *pixel = RGB8 {
            r: ((x * 255) / width) as u8,
            g,
            b: ((y * 255) / height) as u8,
        };
    }

    let encoded = PngEncoder::default().encode(img).unwrap();
    let mut f = std::fs::File::create("frac.png").unwrap();
    f.write_all(&encoded[..]).unwrap();
}

(you can find this code in the examples)

disjoint mutable views example

one of the coolest features of picture is the ability to have disjoint mutable views into a view. the following example swaps the two horizontal halves of an ImgBuf and then saves the result.

use picture::{
    formats::png::{PngDecoder, PngEncoder, PngImage},
    prelude::*,
};
use std::io::Write;

fn swap<I1, I2, P>(a: &mut I1, b: &mut I2)
where
    I1: ImgViewMut<Pixel = P>,
    I2: ImgViewMut<Pixel = P>,
{
    assert!(a.dimensions() == b.dimensions());
    a.pixels_mut()
        .zip(b.pixels_mut())
        .for_each(|(a, b)| std::mem::swap(a, b));
}

fn main() {
    let colorful = PngDecoder
        .decode_from_path("examples/images/colorful.png")
        .unwrap();

    let PngImage::Rgb(mut colorful) = colorful else {
        unreachable!()
    };

    let (mut a, mut b) = colorful.split_x_at_mut(colorful.width() / 2).unwrap();
    swap(&mut a, &mut b);

    let encoded = PngEncoder::default().encode(colorful).unwrap();
    let mut f = std::fs::File::create("swapped.png").unwrap();
    f.write_all(&encoded[..]).unwrap();
}

(you can find this code in the examples as well!)

the ImgViewMut trait has many methods to obtain disjoint mutable views: split_x_at_mut, split_y_at_mut and view_mut_multiple.

performance

i've written a few benchmarks to compare picture and image, but i'm not experienced at writing these - so take these results with a grain of salt. benchmarks were executed on my ryzen 5 1600 with lto = "thin".

Benchmark Time
Diff 256x256/Picture/ [377.45 µs 379.36 µs 381.44 µs]
Diff 256x256/Image/ [735.66 µs 744.57 µs 753.50 µs]
Fractal 256x256/Picture/ [8.2667 ms 8.3979 ms 8.5334 ms]
Fractal 256x256/Image/ [9.0005 ms 9.1315 ms 9.2613 ms]
Match 16x16 in 256x256/Picture/ [183.25 ms 184.11 ms 185.00 ms]
Match 16x16 in 256x256/Image/ [188.99 ms 190.29 ms 191.65 ms]
Lancoz Downsample/Picture/ [54.864 ms 55.451 ms 56.069 ms]
Lancoz Downsample/Image/ [55.418 ms 56.024 ms 56.655 ms]
Lancoz Upsample/Picture/ [396.48 ms 401.48 ms 406.81 ms]
Lancoz Upsample/Image/ [861.12 ms 863.81 ms 866.49 ms]

you can try running these yourself with a simple cargo bench.

unsafe

this crate uses unsafe in many places, but i've tried my best to make it all sound - including adding SAFETY: comments above every unsafe block. it's my first big project with abundancy of unsafe, however, so if you find any soundness holes please let me know!

Dependencies

~1.7–2.3MB
~46K SLoC