0
0
Fork 0
mirror of https://github.com/denoland/deno.git synced 2025-02-01 20:25:12 -05:00
denoland-deno/ext/canvas/image_ops.rs

646 lines
19 KiB
Rust

// Copyright 2018-2024 the Deno authors. All rights reserved. MIT license.
use bytemuck::cast_slice;
use deno_core::error::AnyError;
use image::ColorType;
use image::DynamicImage;
use image::GenericImageView;
use image::ImageBuffer;
use image::Luma;
use image::LumaA;
use image::Pixel;
use image::Primitive;
use image::Rgb;
use image::Rgba;
use lcms2::PixelFormat;
use lcms2::Pod;
use lcms2::Profile;
use lcms2::Transform;
use num_traits::NumCast;
use num_traits::SaturatingMul;
pub(crate) trait PremultiplyAlpha {
fn premultiply_alpha(&self) -> Self;
}
impl<T: Primitive> PremultiplyAlpha for LumaA<T> {
fn premultiply_alpha(&self) -> Self {
let max_t = T::DEFAULT_MAX_VALUE;
let mut pixel = [self.0[0], self.0[1]];
let alpha_index = pixel.len() - 1;
let alpha = pixel[alpha_index];
let normalized_alpha = alpha.to_f32().unwrap() / max_t.to_f32().unwrap();
if normalized_alpha == 0.0 {
return LumaA::<T>([pixel[0], pixel[alpha_index]]);
}
for rgb in pixel.iter_mut().take(alpha_index) {
*rgb = NumCast::from((rgb.to_f32().unwrap() * normalized_alpha).round())
.unwrap()
}
LumaA::<T>([pixel[0], pixel[alpha_index]])
}
}
impl<T: Primitive> PremultiplyAlpha for Rgba<T> {
fn premultiply_alpha(&self) -> Self {
let max_t = T::DEFAULT_MAX_VALUE;
let mut pixel = [self.0[0], self.0[1], self.0[2], self.0[3]];
let alpha_index = pixel.len() - 1;
let alpha = pixel[alpha_index];
let normalized_alpha = alpha.to_f32().unwrap() / max_t.to_f32().unwrap();
if normalized_alpha == 0.0 {
return Rgba::<T>([pixel[0], pixel[1], pixel[2], pixel[alpha_index]]);
}
for rgb in pixel.iter_mut().take(alpha_index) {
*rgb = NumCast::from((rgb.to_f32().unwrap() * normalized_alpha).round())
.unwrap()
}
Rgba::<T>([pixel[0], pixel[1], pixel[2], pixel[alpha_index]])
}
}
// make public if needed
fn process_premultiply_alpha<I, P, S>(image: &I) -> ImageBuffer<P, Vec<S>>
where
I: GenericImageView<Pixel = P>,
P: Pixel<Subpixel = S> + PremultiplyAlpha + 'static,
S: Primitive + 'static,
{
let (width, height) = image.dimensions();
let mut out = ImageBuffer::new(width, height);
for (x, y, pixel) in image.pixels() {
let pixel = pixel.premultiply_alpha();
out.put_pixel(x, y, pixel);
}
out
}
/// Premultiply the alpha channel of the image.
pub(crate) fn premultiply_alpha(
image: DynamicImage,
unmatch_color_handler: fn(
ColorType,
DynamicImage,
) -> Result<DynamicImage, AnyError>,
) -> Result<DynamicImage, AnyError> {
let color = image.color();
match color {
ColorType::La8 => Ok(DynamicImage::ImageLumaA8(process_premultiply_alpha(
&image.to_luma_alpha8(),
))),
ColorType::Rgba8 => Ok(DynamicImage::ImageRgba8(
process_premultiply_alpha(&image.to_rgba8()),
)),
ColorType::La16 => Ok(DynamicImage::ImageLumaA16(
process_premultiply_alpha(&image.to_luma_alpha16()),
)),
ColorType::Rgba16 => Ok(DynamicImage::ImageRgba16(
process_premultiply_alpha(&image.to_rgba16()),
)),
x => unmatch_color_handler(x, image),
}
}
pub(crate) trait UnpremultiplyAlpha {
/// To determine if the image is premultiplied alpha,
/// checking premultiplied RGBA value is one where any of the R/G/B channel values exceeds the alpha channel value.\
/// https://www.w3.org/TR/webgpu/#color-spaces
fn is_premultiplied_alpha(&self) -> bool;
fn unpremultiply_alpha(&self) -> Self;
}
impl<T: Primitive + SaturatingMul + Ord> UnpremultiplyAlpha for Rgba<T> {
fn is_premultiplied_alpha(&self) -> bool {
let max_t = T::DEFAULT_MAX_VALUE;
let pixel = [self.0[0], self.0[1], self.0[2]];
let alpha_index = self.0.len() - 1;
let alpha = self.0[alpha_index];
match pixel.iter().max() {
Some(rgb_max) => rgb_max < &max_t.saturating_mul(&alpha),
// usually doesn't reach here
None => false,
}
}
fn unpremultiply_alpha(&self) -> Self {
let max_t = T::DEFAULT_MAX_VALUE;
let mut pixel = [self.0[0], self.0[1], self.0[2], self.0[3]];
let alpha_index = pixel.len() - 1;
let alpha = pixel[alpha_index];
for rgb in pixel.iter_mut().take(alpha_index) {
*rgb = NumCast::from(
(rgb.to_f32().unwrap()
/ (alpha.to_f32().unwrap() / max_t.to_f32().unwrap()))
.round(),
)
.unwrap();
}
Rgba::<T>([pixel[0], pixel[1], pixel[2], pixel[alpha_index]])
}
}
impl<T: Primitive + SaturatingMul + Ord> UnpremultiplyAlpha for LumaA<T> {
fn is_premultiplied_alpha(&self) -> bool {
let max_t = T::DEFAULT_MAX_VALUE;
let pixel = [self.0[0]];
let alpha_index = self.0.len() - 1;
let alpha = self.0[alpha_index];
pixel[0] < max_t.saturating_mul(&alpha)
}
fn unpremultiply_alpha(&self) -> Self {
let max_t = T::DEFAULT_MAX_VALUE;
let mut pixel = [self.0[0], self.0[1]];
let alpha_index = pixel.len() - 1;
let alpha = pixel[alpha_index];
for rgb in pixel.iter_mut().take(alpha_index) {
*rgb = NumCast::from(
(rgb.to_f32().unwrap()
/ (alpha.to_f32().unwrap() / max_t.to_f32().unwrap()))
.round(),
)
.unwrap();
}
LumaA::<T>([pixel[0], pixel[alpha_index]])
}
}
// make public if needed
fn process_unpremultiply_alpha<I, P, S>(image: &I) -> ImageBuffer<P, Vec<S>>
where
I: GenericImageView<Pixel = P>,
P: Pixel<Subpixel = S> + UnpremultiplyAlpha + 'static,
S: Primitive + 'static,
{
let (width, height) = image.dimensions();
let mut out = ImageBuffer::new(width, height);
let is_premultiplied_alpha = image
.pixels()
.any(|(_, _, pixel)| pixel.is_premultiplied_alpha());
for (x, y, pixel) in image.pixels() {
let pixel = if is_premultiplied_alpha {
pixel.unpremultiply_alpha()
} else {
// return the original
pixel
};
out.put_pixel(x, y, pixel);
}
out
}
/// Invert the premultiplied alpha channel of the image.
pub(crate) fn unpremultiply_alpha(
image: DynamicImage,
unmatch_color_handler: fn(
ColorType,
DynamicImage,
) -> Result<DynamicImage, AnyError>,
) -> Result<DynamicImage, AnyError> {
match image.color() {
ColorType::La8 => Ok(DynamicImage::ImageLumaA8(
process_unpremultiply_alpha(&image.to_luma_alpha8()),
)),
ColorType::Rgba8 => Ok(DynamicImage::ImageRgba8(
process_unpremultiply_alpha(&image.to_rgba8()),
)),
ColorType::La16 => Ok(DynamicImage::ImageLumaA16(
process_unpremultiply_alpha(&image.to_luma_alpha16()),
)),
ColorType::Rgba16 => Ok(DynamicImage::ImageRgba16(
process_unpremultiply_alpha(&image.to_rgba16()),
)),
x => unmatch_color_handler(x, image),
}
}
pub(crate) trait SliceToPixel {
fn slice_to_pixel(pixel: &[u8]) -> Self;
}
impl<T: Primitive + Pod> SliceToPixel for Luma<T> {
fn slice_to_pixel(pixel: &[u8]) -> Self {
let pixel: &[T] = cast_slice(pixel);
let pixel = [pixel[0]];
Luma::<T>(pixel)
}
}
impl<T: Primitive + Pod> SliceToPixel for LumaA<T> {
fn slice_to_pixel(pixel: &[u8]) -> Self {
let pixel: &[T] = cast_slice(pixel);
let pixel = [pixel[0], pixel[1]];
LumaA::<T>(pixel)
}
}
impl<T: Primitive + Pod> SliceToPixel for Rgb<T> {
fn slice_to_pixel(pixel: &[u8]) -> Self {
let pixel: &[T] = cast_slice(pixel);
let pixel = [pixel[0], pixel[1], pixel[2]];
Rgb::<T>(pixel)
}
}
impl<T: Primitive + Pod> SliceToPixel for Rgba<T> {
fn slice_to_pixel(pixel: &[u8]) -> Self {
let pixel: &[T] = cast_slice(pixel);
let pixel = [pixel[0], pixel[1], pixel[2], pixel[3]];
Rgba::<T>(pixel)
}
}
/// Convert the pixel slice to an array to avoid the copy to Vec.
/// I implemented this trait because of I couldn't find a way to effectively combine
/// the `Transform` struct of `lcms2` and `Pixel` trait of `image`.
/// If there is an implementation that is safer and can withstand changes, I would like to adopt it.
pub(crate) trait SliceToArray<const N: usize> {
fn slice_to_array(pixel: &[u8]) -> [u8; N];
}
macro_rules! impl_slice_to_array {
($type:ty, $n:expr) => {
impl<T: Primitive + Pod> SliceToArray<$n> for $type {
fn slice_to_array(pixel: &[u8]) -> [u8; $n] {
let mut dst = [0_u8; $n];
dst.copy_from_slice(&pixel[..$n]);
dst
}
}
};
}
impl_slice_to_array!(Luma<T>, 1);
impl_slice_to_array!(Luma<T>, 2);
impl_slice_to_array!(LumaA<T>, 2);
impl_slice_to_array!(LumaA<T>, 4);
impl_slice_to_array!(Rgb<T>, 3);
impl_slice_to_array!(Rgb<T>, 6);
impl_slice_to_array!(Rgba<T>, 4);
impl_slice_to_array!(Rgba<T>, 8);
// make public if needed
fn process_icc_profile_conversion<P, S, const N: usize>(
image: &DynamicImage,
input_icc_profile: Profile,
output_icc_profile: Profile,
) -> ImageBuffer<P, Vec<S>>
where
P: Pixel<Subpixel = S> + SliceToPixel + SliceToArray<N> + 'static,
S: Primitive + 'static,
{
let (width, height) = image.dimensions();
let mut out = ImageBuffer::new(width, height);
let chunk_size = image.color().bytes_per_pixel() as usize;
let pixel_iter = image
.as_bytes()
.chunks_exact(chunk_size)
.zip(image.pixels());
let pixel_format = match image.color() {
ColorType::L8 => PixelFormat::GRAY_8,
ColorType::L16 => PixelFormat::GRAY_16,
ColorType::La8 => PixelFormat::GRAYA_8,
ColorType::La16 => PixelFormat::GRAYA_16,
ColorType::Rgb8 => PixelFormat::RGB_8,
ColorType::Rgb16 => PixelFormat::RGB_16,
ColorType::Rgba8 => PixelFormat::RGBA_8,
ColorType::Rgba16 => PixelFormat::RGBA_16,
// This arm usually doesn't reach, but it should be handled with returning the original image.
_ => {
return {
for (pixel, (x, y, _)) in pixel_iter {
out.put_pixel(x, y, P::slice_to_pixel(pixel));
}
out
}
}
};
let transformer = Transform::new(
&input_icc_profile,
pixel_format,
&output_icc_profile,
pixel_format,
output_icc_profile.header_rendering_intent(),
);
for (pixel, (x, y, _)) in pixel_iter {
let pixel = match transformer {
Ok(ref transformer) => {
let mut dst = P::slice_to_array(pixel);
transformer.transform_in_place(&mut dst);
dst
}
// This arm will reach when the ffi call fails.
Err(_) => P::slice_to_array(pixel),
};
out.put_pixel(x, y, P::slice_to_pixel(&pixel));
}
out
}
#[rustfmt::skip]
/// Convert the color space of the image from the ICC profile to sRGB.
pub(crate) fn to_srgb_from_icc_profile(
image: DynamicImage,
icc_profile: Option<Vec<u8>>,
unmatch_color_handler: fn(ColorType, DynamicImage) -> Result<DynamicImage, AnyError>,
) -> Result<DynamicImage, AnyError> {
match icc_profile {
// If there is no color profile information, return the image as is.
None => Ok(image),
Some(icc_profile) => match Profile::new_icc(&icc_profile) {
// If the color profile information is invalid, return the image as is.
Err(_) => Ok(image),
Ok(icc_profile) => {
let srgb_icc_profile = Profile::new_srgb();
match image.color() {
ColorType::L8 => {
Ok(DynamicImage::ImageLuma8(process_icc_profile_conversion::<_,_,1>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::L16 => {
Ok(DynamicImage::ImageLuma16(process_icc_profile_conversion::<_,_,2>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::La8 => {
Ok(DynamicImage::ImageLumaA8(process_icc_profile_conversion::<_,_,2>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::La16 => {
Ok(DynamicImage::ImageLumaA16(process_icc_profile_conversion::<_, _, 4>(&image,icc_profile,srgb_icc_profile)))
},
ColorType::Rgb8 => {
Ok(DynamicImage::ImageRgb8(process_icc_profile_conversion::<_,_,3>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::Rgb16 => {
Ok(DynamicImage::ImageRgb16(process_icc_profile_conversion::<_,_,6>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::Rgba8 => {
Ok(DynamicImage::ImageRgba8(process_icc_profile_conversion::<_,_,4>(&image,icc_profile,srgb_icc_profile)))
}
ColorType::Rgba16 => {
Ok(DynamicImage::ImageRgba16(process_icc_profile_conversion::<_,_,8>(&image,icc_profile,srgb_icc_profile)))
}
x => unmatch_color_handler(x, image),
}
}
},
}
}
// NOTE: The following code is not used in the current implementation,
// but it is left as a reference for future use about implementing CanvasRenderingContext2D.
// https://github.com/denoland/deno/issues/5701#issuecomment-1833304511
// // reference
// // https://www.w3.org/TR/css-color-4/#color-conversion-code
// fn srgb_to_linear<T: Primitive>(value: T) -> f32 {
// if value.to_f32().unwrap() <= 0.04045 {
// value.to_f32().unwrap() / 12.92
// } else {
// ((value.to_f32().unwrap() + 0.055) / 1.055).powf(2.4)
// }
// }
// // reference
// // https://www.w3.org/TR/css-color-4/#color-conversion-code
// fn linear_to_display_p3<T: Primitive>(value: T) -> f32 {
// if value.to_f32().unwrap() <= 0.0031308 {
// value.to_f32().unwrap() * 12.92
// } else {
// 1.055 * value.to_f32().unwrap().powf(1.0 / 2.4) - 0.055
// }
// }
// fn normalize_value_to_0_1<T: Primitive>(value: T) -> f32 {
// value.to_f32().unwrap() / T::DEFAULT_MAX_VALUE.to_f32().unwrap()
// }
// fn unnormalize_value_from_0_1<T: Primitive>(value: f32) -> T {
// NumCast::from(
// (value.clamp(0.0, 1.0) * T::DEFAULT_MAX_VALUE.to_f32().unwrap()).round(),
// )
// .unwrap()
// }
// fn apply_conversion_matrix_srgb_to_display_p3<T: Primitive>(
// r: T,
// g: T,
// b: T,
// ) -> (T, T, T) {
// // normalize the value to 0.0 - 1.0
// let (r, g, b) = (
// normalize_value_to_0_1(r),
// normalize_value_to_0_1(g),
// normalize_value_to_0_1(b),
// );
// // sRGB -> Linear RGB
// let (r, g, b) = (srgb_to_linear(r), srgb_to_linear(g), srgb_to_linear(b));
// // Display-P3 (RGB) -> Display-P3 (XYZ)
// //
// // inv[ P3-D65 (D65) to XYZ ] * [ sRGB (D65) to XYZ ]
// // http://www.brucelindbloom.com/index.html?Eqn_RGB_XYZ_Matrix.html
// // https://fujiwaratko.sakura.ne.jp/infosci/colorspace/colorspace2_e.html
// // [ sRGB (D65) to XYZ ]
// #[rustfmt::skip]
// let (m1x, m1y, m1z) = (
// [0.4124564, 0.3575761, 0.1804375],
// [0.2126729, 0.7151522, 0.0721750],
// [0.0193339, 0.119_192, 0.9503041],
// );
// let (r, g, b) = (
// r * m1x[0] + g * m1x[1] + b * m1x[2],
// r * m1y[0] + g * m1y[1] + b * m1y[2],
// r * m1z[0] + g * m1z[1] + b * m1z[2],
// );
// // inv[ P3-D65 (D65) to XYZ ]
// #[rustfmt::skip]
// let (m2x, m2y, m2z) = (
// [ 2.493_497, -0.931_383_6, -0.402_710_8 ],
// [ -0.829_489, 1.762_664_1, 0.023_624_687 ],
// [ 0.035_845_83, -0.076_172_39, 0.956_884_5 ],
// );
// let (r, g, b) = (
// r * m2x[0] + g * m2x[1] + b * m2x[2],
// r * m2y[0] + g * m2y[1] + b * m2y[2],
// r * m2z[0] + g * m2z[1] + b * m2z[2],
// );
// // This calculation is similar as above that it is a little faster, but less accurate.
// // let r = 0.8225 * r + 0.1774 * g + 0.0000 * b;
// // let g = 0.0332 * r + 0.9669 * g + 0.0000 * b;
// // let b = 0.0171 * r + 0.0724 * g + 0.9108 * b;
// // Display-P3 (Linear) -> Display-P3
// let (r, g, b) = (
// linear_to_display_p3(r),
// linear_to_display_p3(g),
// linear_to_display_p3(b),
// );
// // unnormalize the value from 0.0 - 1.0
// (
// unnormalize_value_from_0_1(r),
// unnormalize_value_from_0_1(g),
// unnormalize_value_from_0_1(b),
// )
// }
// trait ColorSpaceConversion {
// /// Display P3 Color Encoding (v 1.0)
// /// https://www.color.org/chardata/rgb/DisplayP3.xalter
// fn srgb_to_display_p3(&self) -> Self;
// }
// impl<T: Primitive> ColorSpaceConversion for Rgb<T> {
// fn srgb_to_display_p3(&self) -> Self {
// let (r, g, b) = (self.0[0], self.0[1], self.0[2]);
// let (r, g, b) = apply_conversion_matrix_srgb_to_display_p3(r, g, b);
// Rgb::<T>([r, g, b])
// }
// }
// impl<T: Primitive> ColorSpaceConversion for Rgba<T> {
// fn srgb_to_display_p3(&self) -> Self {
// let (r, g, b, a) = (self.0[0], self.0[1], self.0[2], self.0[3]);
// let (r, g, b) = apply_conversion_matrix_srgb_to_display_p3(r, g, b);
// Rgba::<T>([r, g, b, a])
// }
// }
// // make public if needed
// fn process_srgb_to_display_p3<I, P, S>(image: &I) -> ImageBuffer<P, Vec<S>>
// where
// I: GenericImageView<Pixel = P>,
// P: Pixel<Subpixel = S> + ColorSpaceConversion + 'static,
// S: Primitive + 'static,
// {
// let (width, height) = image.dimensions();
// let mut out = ImageBuffer::new(width, height);
// for (x, y, pixel) in image.pixels() {
// let pixel = pixel.srgb_to_display_p3();
// out.put_pixel(x, y, pixel);
// }
// out
// }
// /// Convert the color space of the image from sRGB to Display-P3.
// fn srgb_to_display_p3(
// image: DynamicImage,
// unmatch_color_handler: fn(
// ColorType,
// DynamicImage,
// ) -> Result<DynamicImage, AnyError>,
// ) -> Result<DynamicImage, AnyError> {
// match image.color() {
// // The conversion of the lumincance color types to the display-p3 color space is meaningless.
// ColorType::L8 => Ok(DynamicImage::ImageLuma8(image.to_luma8())),
// ColorType::L16 => Ok(DynamicImage::ImageLuma16(image.to_luma16())),
// ColorType::La8 => Ok(DynamicImage::ImageLumaA8(image.to_luma_alpha8())),
// ColorType::La16 => Ok(DynamicImage::ImageLumaA16(image.to_luma_alpha16())),
// ColorType::Rgb8 => Ok(DynamicImage::ImageRgb8(process_srgb_to_display_p3(
// &image.to_rgb8(),
// ))),
// ColorType::Rgb16 => Ok(DynamicImage::ImageRgb16(
// process_srgb_to_display_p3(&image.to_rgb16()),
// )),
// ColorType::Rgba8 => Ok(DynamicImage::ImageRgba8(
// process_srgb_to_display_p3(&image.to_rgba8()),
// )),
// ColorType::Rgba16 => Ok(DynamicImage::ImageRgba16(
// process_srgb_to_display_p3(&image.to_rgba16()),
// )),
// x => unmatch_color_handler(x, image),
// }
// }
#[cfg(test)]
mod tests {
use super::*;
use image::Rgba;
#[test]
fn test_premultiply_alpha() {
let rgba = Rgba::<u8>([255, 128, 0, 128]);
let rgba = rgba.premultiply_alpha();
assert_eq!(rgba, Rgba::<u8>([128, 64, 0, 128]));
let rgba = Rgba::<u8>([255, 255, 255, 255]);
let rgba = rgba.premultiply_alpha();
assert_eq!(rgba, Rgba::<u8>([255, 255, 255, 255]));
}
#[test]
fn test_unpremultiply_alpha() {
let rgba = Rgba::<u8>([127, 0, 0, 127]);
let rgba = rgba.unpremultiply_alpha();
assert_eq!(rgba, Rgba::<u8>([255, 0, 0, 127]));
}
// #[test]
// fn test_apply_conversion_matrix_srgb_to_display_p3() {
// let (r, g, b) = apply_conversion_matrix_srgb_to_display_p3(255_u8, 0, 0);
// assert_eq!(r, 234);
// assert_eq!(g, 51);
// assert_eq!(b, 35);
// let (r, g, b) = apply_conversion_matrix_srgb_to_display_p3(0_u8, 255, 0);
// assert_eq!(r, 117);
// assert_eq!(g, 251);
// assert_eq!(b, 76);
// let (r, g, b) = apply_conversion_matrix_srgb_to_display_p3(0_u8, 0, 255);
// assert_eq!(r, 0);
// assert_eq!(g, 0);
// assert_eq!(b, 245);
// let (r, g, b) =
// apply_conversion_matrix_srgb_to_display_p3(255_u8, 255, 255);
// assert_eq!(r, 255);
// assert_eq!(g, 255);
// assert_eq!(b, 255);
// }
}