DCV Color Primitives - dcp

DCV Color Primitives is a library to perform image color model conversion.

Design guidelines

• Aware of the underlying hardware and supplemental cpu extension sets (up to avx2)
• Support data coming from a single buffer or coming from multiple image planes
• Support non-tightly packed data
• Support images greater than 4GB (64 bit)
• Support ARM (aarch64)[*]
• Support WebAssembly[*]

[*]: Supplemental cpu extension sets not yet supported.

Image format conversion

The library is currenty able to convert the following pixel formats:

Color models

The supported color models are:

• YCbCr, ITU-R Recommendation BT.601 (standard video system)
• YCbCr, ITU-R Recommendation BT.709 (CSC systems)

Both standard range (0-235) and full range (0-255) are supported.

Requirements

• Rust 1.70 and newer

Windows

• Install rustup: https://www.rust-lang.org/tools/install

Linux

• Install rustup (see https://forge.rust-lang.org/infra/other-installation-methods.html)

  curl https://sh.rustup.rs -sSf | sh
  

You may require administrative privileges.

Building

Open a terminal inside the library root directory.

To build for debug experience:

cargo build

To build an optimized library:

cargo build --release

Run unit tests:

cargo test

Run benchmark:

cargo bench

Advanced benchmark mode. There are two benchmark scripts:

• run-bench.ps1 for Windows
• run-bench.sh for Linux and MacOS

They allow to obtain more stable results than cargo bench, by reducing variance due to:

• CPU migration
• File system caching
• Process priority

Moreover, the Linux script support hardware performance counters, e.g. it is possible to output consumed CPU cycles instead of elapsed time.

Linux examples:

./run-bench -c 1 # runs cargo bench and outputs CPU cycles
./run.bench -c 1 -p "/i420" # runs cargo bench, output CPU cycles, filtering tests that contains '/i420'

WebAssembly

Install the needed dependencies:

rustup target add wasm32-unknown-unknown

To build for debug experience:

cargo build --target wasm32-unknown-unknown

To test, ensure you have installed wasm-pack. Then:

wasm-pack test --node

Usage

Image conversion

Convert an image from bgra to nv12 (single plane) format containing yuv in BT601:

use dcv_color_primitives as dcp;
use dcp::{convert_image, ColorSpace, ImageFormat, PixelFormat};

fn main() {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;

    let src_data = Box::new([0u8; 4 * (WIDTH as usize) * (HEIGHT as usize)]);
    let mut dst_data = Box::new([0u8; 3 * (WIDTH as usize) * (HEIGHT as usize) / 2]);

    let src_format = ImageFormat {
        pixel_format: PixelFormat::Bgra,
        color_space: ColorSpace::Rgb,
        num_planes: 1,
    };

    let dst_format = ImageFormat {
        pixel_format: PixelFormat::Nv12,
        color_space: ColorSpace::Bt601,
        num_planes: 1,
    };

    convert_image(
        WIDTH,
        HEIGHT,
        &src_format,
        None,
        &[&*src_data],
        &dst_format,
        None,
        &mut [&mut *dst_data],
    );
}

Error Handling

The library functions return a Result describing the operation outcome:

In the following example, result will match Err(ErrorKind::InvalidValue), because ColorSpace::Bt709 color space is not compatible with PixelFormat::Bgra:

use dcv_color_primitives as dcp;
use dcp::{convert_image, ColorSpace, ErrorKind, ImageFormat, PixelFormat};

fn main() {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;

    let src_data = Box::new([0u8; 4 * (WIDTH as usize) * (HEIGHT as usize)]);
    let mut dst_data = Box::new([0u8; 3 * (WIDTH as usize) * (HEIGHT as usize) / 2]);

    let src_format = ImageFormat {
        pixel_format: PixelFormat::Bgra,
        color_space: ColorSpace::Bt709,
        num_planes: 1,
    };

    let dst_format = ImageFormat {
        pixel_format: PixelFormat::Nv12,
        color_space: ColorSpace::Bt601,
        num_planes: 1,
    };

    let status = convert_image(
        WIDTH,
        HEIGHT,
        &src_format,
        None,
        &[&*src_data],
        &dst_format,
        None,
        &mut [&mut *dst_data],
    );

    match status {
        Err(ErrorKind::InvalidValue) => (),
        _ => panic!("Expected ErrorKind::InvalidValue"),
    }
}

Even better, you might want to propagate errors to the caller function or mix with some other error types:

use dcv_color_primitives as dcp;
use dcp::{convert_image, ColorSpace, ImageFormat, PixelFormat};
use std::error;

fn main() -> Result<(), Box<dyn error::Error>> {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;

    let src_data = Box::new([0u8; 4 * (WIDTH as usize) * (HEIGHT as usize)]);
    let mut dst_data = Box::new([0u8; 3 * (WIDTH as usize) * (HEIGHT as usize) / 2]);

    let src_format = ImageFormat {
        pixel_format: PixelFormat::Bgra,
        color_space: ColorSpace::Bt709,
        num_planes: 1,
    };

    let dst_format = ImageFormat {
        pixel_format: PixelFormat::Nv12,
        color_space: ColorSpace::Bt601,
        num_planes: 1,
    };

    convert_image(
        WIDTH,
        HEIGHT,
        &src_format,
        None,
        &[&*src_data],
        &dst_format,
        None,
        &mut [&mut *dst_data],
    )?;

    Ok(())
}

Buffer size computation

So far, buffers were sized taking into account the image pixel format and dimensions; However, you can use a function to compute how many bytes are needed to store an image of a given format and size:

use dcv_color_primitives as dcp;
use dcp::{get_buffers_size, ColorSpace, ImageFormat, PixelFormat};
use std::error;

fn main() -> Result<(), Box<dyn error::Error>> {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;
    const NUM_PLANES: u32 = 1;

    let format = ImageFormat {
        pixel_format: PixelFormat::Bgra,
        color_space: ColorSpace::Rgb,
        num_planes: NUM_PLANES,
    };

    let sizes: &mut [usize] = &mut [0usize; NUM_PLANES as usize];
    get_buffers_size(WIDTH, HEIGHT, &format, None, sizes)?;

    let buffer: Vec<_> = vec![0u8; sizes[0]];

    // Do something with buffer
    // --snip--

    Ok(())
}

Image planes

If your data is scattered in multiple buffers that are not necessarily contiguous, you can provide image planes:

use dcv_color_primitives as dcp;
use dcp::{convert_image, get_buffers_size, ColorSpace, ImageFormat, PixelFormat};
use std::error;

fn main() -> Result<(), Box<dyn error::Error>> {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;
    const NUM_SRC_PLANES: u32 = 2;
    const NUM_DST_PLANES: u32 = 1;

    let src_format = ImageFormat {
        pixel_format: PixelFormat::Nv12,
        color_space: ColorSpace::Bt709,
        num_planes: NUM_SRC_PLANES,
    };

    let src_sizes: &mut [usize] = &mut [0usize; NUM_SRC_PLANES as usize];
    get_buffers_size(WIDTH, HEIGHT, &src_format, None, src_sizes)?;

    let src_y: Vec<_> = vec![0u8; src_sizes[0]];
    let src_uv: Vec<_> = vec![0u8; src_sizes[1]];

    let dst_format = ImageFormat {
        pixel_format: PixelFormat::Bgra,
        color_space: ColorSpace::Rgb,
        num_planes: NUM_DST_PLANES,
    };

    let dst_sizes: &mut [usize] = &mut [0usize; NUM_DST_PLANES as usize];
    get_buffers_size(WIDTH, HEIGHT, &dst_format, None, dst_sizes)?;

    let mut dst_rgba: Vec<_> = vec![0u8; dst_sizes[0]];

    convert_image(
        WIDTH,
        HEIGHT,
        &src_format,
        None,
        &[&src_y[..], &src_uv[..]],
        &dst_format,
        None,
        &mut [&mut dst_rgba[..]],
    )?;

    Ok(())
}

Stride support

To take into account data which is not tightly packed, you can provide image strides:

use dcv_color_primitives as dcp;
use dcp::{convert_image, get_buffers_size, ColorSpace, ImageFormat, PixelFormat};
use std::error;

fn main() -> Result<(), Box<dyn error::Error>> {
    const WIDTH: u32 = 640;
    const HEIGHT: u32 = 480;
    const NUM_SRC_PLANES: u32 = 1;
    const NUM_DST_PLANES: u32 = 2;
    const RGB_STRIDE: usize = 4 * (((3 * (WIDTH as usize)) + 3) / 4);

    let src_format = ImageFormat {
        pixel_format: PixelFormat::Bgr,
        color_space: ColorSpace::Rgb,
        num_planes: NUM_SRC_PLANES,
    };

    let src_strides: &[usize] = &[RGB_STRIDE];

    let src_sizes: &mut [usize] = &mut [0usize; NUM_SRC_PLANES as usize];
    get_buffers_size(WIDTH, HEIGHT, &src_format, Some(src_strides), src_sizes)?;

    let src_rgba: Vec<_> = vec![0u8; src_sizes[0]];

    let dst_format = ImageFormat {
        pixel_format: PixelFormat::Nv12,
        color_space: ColorSpace::Bt709,
        num_planes: NUM_DST_PLANES,
    };

    let dst_sizes: &mut [usize] = &mut [0usize; NUM_DST_PLANES as usize];
    get_buffers_size(WIDTH, HEIGHT, &dst_format, None, dst_sizes)?;

    let mut dst_y: Vec<_> = vec![0u8; dst_sizes[0]];
    let mut dst_uv: Vec<_> = vec![0u8; dst_sizes[1]];

    convert_image(
        WIDTH,
        HEIGHT,
        &src_format,
        Some(src_strides),
        &[&src_rgba[..]],
        &dst_format,
        None,
        &mut [&mut dst_y[..], &mut dst_uv[..]],
    )?;

    Ok(())
}

See documentation for further information.

C bindings

DCV Color Primitives provides C bindings. A static library will be automatically generated for the default build.

In order to include DCV Color Primitives inside your application library, you need to:

• Statically link to dcv_color_primitives
• Link to ws2_32.lib, userenv.lib, bcrypt.lib and ntdll.lib, for Windows
• Link to libdl and libm, for Linux

The API is slightly different than the rust one. Check dcv_color_primitives.h for examples and further information.

A meson build system is provided in order to build the static library and install it together with include file and a pkgconfig file. There are also some unit tests written in C, to add some coverage also for the bindings. Minimal instructions are provided below, refer to meson's help for further instructions:

• Windows Visual Studio is required. At least the following packages are required:

  • MSBuild
  • MSVC - C++ build tools
  • Windows 10 SDK

  Install meson, you can choose one of the following methods:

  1. Using meson msi installer
  • Download from https://github.com/mesonbuild/meson/releases
  • Install both Meson and Ninja
  2. Install meson through pip
  • Download and install python3: https://www.python.org/downloads/
  • Install meson and ninja:

    pip install meson ninja
    

  Note: Minimum required meson version is 1.0.0.

  All build commands have to be issued from Native Tools Command Prompt for VS (x86 or x64 depending on what platform you want to build)

• Linux The following example is for Ubuntu:

  #install python3
  apt install python3
  
  #install meson. See https://mesonbuild.com/Getting-meson.html for details or if you want to install through pip.
  apt install meson
  
  #install ninja
  apt install ninja-build
  

  You may require administrative privileges.

• Build Move inside the library root directory:

  cd `dcv_color_primitives_root_dir`
  

  Then:

  meson setup --buildtype release builddir
  ninja -C builddir
  

• Run the tests

  cd builddir
  meson test -t 10
  

  A timeout scale factor of 10 is required because some tests take longer than default 30 seconds to complete.

• Install

  ninja -C builddir install
  

License

This library is licensed under the MIT-0 License. See the LICENSE file.