Cykooz / fast_image_resize

Rust library for fast image resizing with using of SIMD instructions.
Apache License 2.0
312 stars 29 forks source link
image resize resize-image rust-lang

fast_image_resize

github crates.io docs.rs

Rust library for fast image resizing with using of SIMD instructions.

CHANGELOG

Supported pixel formats and available optimizations:

Format Description SSE4.1 AVX2 Neon Wasm32 SIMD128
U8 One u8 component per pixel (e.g. L) + + + +
U8x2 Two u8 components per pixel (e.g. LA) + + + +
U8x3 Three u8 components per pixel (e.g. RGB) + + + +
U8x4 Four u8 components per pixel (e.g. RGBA, RGBx, CMYK) + + + +
U16 One u16 components per pixel (e.g. L16) + + + +
U16x2 Two u16 components per pixel (e.g. LA16) + + + +
U16x3 Three u16 components per pixel (e.g. RGB16) + + + +
U16x4 Four u16 components per pixel (e.g. RGBA16, RGBx16, CMYK16) + + + +
I32 One i32 component per pixel (e.g. L32) - - - -
F32 One f32 component per pixel (e.g. L32F) + + - -
F32x2 Two f32 components per pixel (e.g. LA32F) + + - -
F32x3 Three f32 components per pixel (e.g. RGB32F) + + - -
F32x4 Four f32 components per pixel (e.g. RGBA32F) + + - -

Colorspace

Resizer from this crate does not convert image into linear colorspace during a resize process. If it is important for you to resize images with a non-linear color space (e.g. sRGB) correctly, then you have to convert it to a linear color space before resizing and convert back to the color space of result image. Read more about resizing with respect to color space.

This crate provides the PixelComponentMapper structure that allows you to create colorspace converters for images whose pixels based on u8 and u16 components.

In addition, the crate contains functions create_gamma_22_mapper() and create_srgb_mapper() to create instance of PixelComponentMapper that converts images from sRGB or gamma 2.2 into linear colorspace and back.

Multi-threading

You should enable "rayon" feature to turn on image processing in rayon thread pool.

Some benchmarks in single-threaded mode for x86_64

All benchmarks: _x8664, ARM64, WASM32.

Other libraries used to compare of resizing speed:

Resize RGB8 image (U8x3) 4928x3279 => 852x567

Pipeline:

src_image => resize => dst_image

Nearest Box Bilinear Bicubic Lanczos3
image 34.13 - 88.09 142.99 191.80
resize 8.87 26.97 53.08 98.26 145.74
libvips 2.38 61.57 5.66 9.70 16.03
fir rust 0.28 10.93 15.35 25.77 37.09
fir sse4.1 0.28 3.43 5.39 9.82 15.34
fir avx2 0.28 2.62 3.80 6.89 13.22

Resize RGBA8 image (U8x4) 4928x3279 => 852x567

Pipeline:

src_image => multiply by alpha => resize => divide by alpha => dst_image

Nearest Box Bilinear Bicubic Lanczos3
resize 9.90 37.85 74.20 133.72 201.29
libvips 4.17 169.03 141.52 232.30 330.89
fir rust 0.19 20.66 26.02 37.27 50.21
fir sse4.1 0.19 9.59 11.99 17.79 24.83
fir avx2 0.19 7.21 8.61 13.22 22.41

Resize L8 image (U8) 4928x3279 => 852x567

Pipeline:

src_image => resize => dst_image

Nearest Box Bilinear Bicubic Lanczos3
image 29.07 - 60.25 89.15 117.51
resize 6.42 11.26 20.87 42.87 69.50
libvips 2.57 25.05 6.82 9.85 12.68
fir rust 0.15 4.45 5.57 9.02 12.31
fir sse4.1 0.15 1.52 2.09 3.52 5.65
fir avx2 0.15 1.54 1.76 2.80 4.03

Examples

Resize RGBA8 image

Note: You must enable "image" feature to support of image::DynamicImage. Otherwise, you have to convert such images into supported by the crate image type.

use std::io::BufWriter;

use image::codecs::png::PngEncoder;
use image::{ExtendedColorType, ImageEncoder, ImageReader};

use fast_image_resize::{IntoImageView, Resizer};
use fast_image_resize::images::Image;

fn main() {
    // Read source image from file
    let src_image = ImageReader::open("./data/nasa-4928x3279.png")
        .unwrap()
        .decode()
        .unwrap();

    // Create container for data of destination image
    let dst_width = 1024;
    let dst_height = 768;
    let mut dst_image = Image::new(
        dst_width,
        dst_height,
        src_image.pixel_type().unwrap(),
    );

    // Create Resizer instance and resize source image
    // into buffer of destination image
    let mut resizer = Resizer::new();
    resizer.resize(&src_image, &mut dst_image, None).unwrap();

    // Write destination image as PNG-file
    let mut result_buf = BufWriter::new(Vec::new());
    PngEncoder::new(&mut result_buf)
        .write_image(
            dst_image.buffer(),
            dst_width,
            dst_height,
            src_image.color().into(),
        )
        .unwrap();
}

Resize with cropping

use image::codecs::png::PngEncoder;
use image::{ColorType, ImageReader, GenericImageView};

use fast_image_resize::{IntoImageView, Resizer, ResizeOptions};
use fast_image_resize::images::Image;

fn main() {
    let img = ImageReader::open("./data/nasa-4928x3279.png")
        .unwrap()
        .decode()
        .unwrap();

    // Create container for data of destination image
    let mut dst_image = Image::new(
        1024,
        768,
        img.pixel_type().unwrap(),
    );

    // Create Resizer instance and resize cropped source image
    // into buffer of destination image
    let mut resizer = Resizer::new();
    resizer.resize(
        &img,
        &mut dst_image,
        &ResizeOptions::new().crop(
            10.0,   // left 
            10.0,   // top
            2000.0, // width
            2000.0, // height
        ),
    ).unwrap();
}

Change CPU extensions used by resizer

use fast_image_resize as fr;

fn main() {
    let mut resizer = fr::Resizer::new();
    #[cfg(target_arch = "x86_64")]
    unsafe {
        resizer.set_cpu_extensions(fr::CpuExtensions::Sse4_1);
    }
}