ndarray_vision/core/

colour_models.rs

use crate::core::traits::*;
use crate::core::*;
use ndarray::{prelude::*, s, Data, Zip};
use num_traits::cast::{FromPrimitive, NumCast};
use num_traits::{Num, NumAssignOps};
use std::convert::From;
use std::fmt::Display;

/// Grayscale image
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Gray;
/// RGB colour as intended by sRGB and standardised in IEC 61966-2-1:1999
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct RGB;
/// RGB colour similar to `RGB` type but with an additional channel for alpha
/// transparency
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct RGBA;
/// Hue Saturation Value image
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct HSV;
/// Hue Saturation Intensity image
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct HSI;
/// Hue Saturation Lightness image
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct HSL;
/// YCrCb represents an image as luma, red-difference chroma and blue-difference
/// chroma.
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct YCrCb;
/// CIE XYZ standard - assuming a D50 reference white
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct CIEXYZ;
/// CIE LAB (also known as CIE L*a*b* or Lab) a colour model that represents
/// colour as lightness, and a* and b* as the green-red and blue-yellow colour
/// differences respectively. It is designed to be representative of human
/// perception of colour
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct CIELAB;
/// Similar to `CIELAB` but has a different representation of colour.
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct CIELUV;
/// A single channel image with no colour model specified
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Generic1;
/// A two channel image with no colour model specified
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Generic2;
/// A three channel image with no colour model specified
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Generic3;
/// A four channel image with no colour model specified
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Generic4;
/// A five channel image with no colour model specified
#[derive(Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash, Debug)]
pub struct Generic5;

/// ColourModel trait, this trait reports base parameters for different colour
/// models
pub trait ColourModel {
    /// Number of colour channels for a type.
    fn channels() -> usize {
        3
    }
}

impl RGB {
    /// Remove the gamma from a normalised channel
    pub fn remove_gamma(v: f64) -> f64 {
        if v < 0.04045 {
            v / 12.92
        } else {
            ((v + 0.055) / 1.055).powf(2.4)
        }
    }

    /// Apply the gamma to a normalised channel
    pub fn apply_gamma(v: f64) -> f64 {
        if v < 0.0031308 {
            v * 12.92
        } else {
            1.055 * v.powf(1.0 / 2.4) - 0.055
        }
    }
}

fn rescale_pixel<T>(x: f64) -> T
where
    T: FromPrimitive + Num + NumCast + PixelBound + Display,
{
    let tmax = T::max_pixel().to_f64().unwrap_or(0.0f64);
    let tmin = T::min_pixel().to_f64().unwrap_or(0.0f64);

    let x = x * (tmax - tmin) + tmin;

    T::from_f64(x).unwrap_or_else(T::zero)
}

/// Converts an RGB pixel to a HSV pixel
pub fn rgb_to_hsv<T>(r: T, g: T, b: T) -> (T, T, T)
where
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    let r_norm = normalise_pixel_value(r);
    let g_norm = normalise_pixel_value(g);
    let b_norm = normalise_pixel_value(b);
    let cmax = r_norm.max(g_norm.max(b_norm));
    let cmin = r_norm.min(g_norm.min(b_norm));
    let delta = cmax - cmin;

    let sat = if cmax > 0.0f64 { delta / cmax } else { 0.0f64 };

    let hue = if delta < std::f64::EPSILON {
        0.0 // hue is undefined for full black full white
    } else if cmax <= r_norm {
        60.0 * (((g_norm - b_norm) / delta) % 6.0)
    } else if cmax <= g_norm {
        60.0 * ((b_norm - r_norm) / delta + 2.0)
    } else {
        60.0 * ((r_norm - g_norm) / delta + 4.0)
    };
    let hue = hue / 360.0f64;

    let hue = rescale_pixel(hue);
    let sat = rescale_pixel(sat);
    let val = rescale_pixel(cmax);

    (hue, sat, val)
}

/// Converts a HSV pixel to a RGB pixel
pub fn hsv_to_rgb<T>(h: T, s: T, v: T) -> (T, T, T)
where
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    let h_deg = normalise_pixel_value(h) * 360.0f64;
    let s_norm = normalise_pixel_value(s);
    let v_norm = normalise_pixel_value(v);

    let c = v_norm * s_norm;
    let x = c * (1.0f64 - ((h_deg / 60.0f64) % 2.0f64 - 1.0f64).abs());
    let m = v_norm - c;

    let rgb = if (0.0f64..60.0f64).contains(&h_deg) {
        (c, x, 0.0f64)
    } else if (60.0f64..120.0f64).contains(&h_deg) {
        (x, c, 0.0f64)
    } else if (120.0f64..180.0f64).contains(&h_deg) {
        (0.0f64, c, x)
    } else if (180.0f64..240.0f64).contains(&h_deg) {
        (0.0f64, x, c)
    } else if (240.0f64..300.0f64).contains(&h_deg) {
        (x, 0.0f64, c)
    } else if (300.0f64..360.0f64).contains(&h_deg) {
        (c, 0.0f64, x)
    } else {
        (0.0f64, 0.0f64, 0.0f64)
    };

    let red = rescale_pixel(rgb.0 + m);
    let green = rescale_pixel(rgb.1 + m);
    let blue = rescale_pixel(rgb.2 + m);

    (red, green, blue)
}

impl<U, T> From<ImageBase<U, RGB>> for Image<T, HSV>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, RGB>) -> Self {
        let mut res = Array3::<_>::zeros((image.rows(), image.cols(), HSV::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let red = pix[[0, 0, 0]];
            let green = pix[[0, 0, 1]];
            let blue = pix[[0, 0, 2]];

            let (hue, sat, val) = rgb_to_hsv(red, green, blue);
            res.slice_mut(s![i, j, ..]).assign(&arr1(&[hue, sat, val]));
        });
        Self::from_data(res)
    }
}

impl<T, U> From<ImageBase<U, HSV>> for Image<T, RGB>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, HSV>) -> Self {
        let mut res = Array3::<T>::zeros((image.rows(), image.cols(), RGB::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let h = pix[[0, 0, 0]];
            let s = pix[[0, 0, 1]];
            let v = pix[[0, 0, 2]];

            let (r, g, b) = hsv_to_rgb(h, s, v);
            res.slice_mut(s![i, j, ..]).assign(&arr1(&[r, g, b]));
        });
        Self::from_data(res)
    }
}

impl<T, U> From<ImageBase<U, RGB>> for Image<T, Gray>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, RGB>) -> Self {
        let mut res = Array3::<T>::zeros((image.rows(), image.cols(), Gray::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let r = normalise_pixel_value(pix[[0, 0, 0]]);
            let g = normalise_pixel_value(pix[[0, 0, 1]]);
            let b = normalise_pixel_value(pix[[0, 0, 2]]);

            let gray = (0.3 * r) + (0.59 * g) + (0.11 * b);
            let gray = rescale_pixel(gray);

            res.slice_mut(s![i, j, ..]).assign(&arr1(&[gray]));
        });
        Self::from_data(res)
    }
}

impl<T, U> From<ImageBase<U, Gray>> for Image<T, RGB>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, Gray>) -> Self {
        let mut res = Array3::<T>::zeros((image.rows(), image.cols(), RGB::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let gray = pix[[0, 0, 0]];

            res.slice_mut(s![i, j, ..])
                .assign(&arr1(&[gray, gray, gray]));
        });
        Self::from_data(res)
    }
}

impl<T, U> From<ImageBase<U, RGB>> for Image<T, CIEXYZ>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, RGB>) -> Self {
        let mut res = Array3::<T>::zeros((image.rows(), image.cols(), CIEXYZ::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        let m = arr2(&[
            [0.4360747, 0.3850649, 0.1430804],
            [0.2225045, 0.7168786, 0.0606169],
            [0.0139322, 0.0971045, 0.7141733],
        ]);

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let pixel = pix
                .index_axis(Axis(0), 0)
                .index_axis(Axis(0), 0)
                .mapv(normalise_pixel_value)
                .mapv(RGB::remove_gamma);

            let pixel = m.dot(&pixel);
            let pixel = pixel.mapv(rescale_pixel);

            res.slice_mut(s![i, j, ..]).assign(&pixel);
        });
        Self::from_data(res)
    }
}

impl<T, U> From<ImageBase<U, CIEXYZ>> for Image<T, RGB>
where
    U: Data<Elem = T>,
    T: Copy
        + Clone
        + FromPrimitive
        + Num
        + NumAssignOps
        + NumCast
        + PartialOrd
        + Display
        + PixelBound,
{
    fn from(image: ImageBase<U, CIEXYZ>) -> Self {
        let mut res = Array3::<T>::zeros((image.rows(), image.cols(), RGB::channels()));
        let window = image.data.windows((1, 1, image.channels()));

        let m = arr2(&[
            [3.1338561, -1.6168667, -0.4906146],
            [-0.9787684, 1.9161415, 0.0334540],
            [0.0719453, -0.2289914, 1.4052427],
        ]);

        Zip::indexed(window).for_each(|(i, j, _), pix| {
            let pixel = pix
                .index_axis(Axis(0), 0)
                .index_axis(Axis(0), 0)
                .mapv(normalise_pixel_value);

            let pixel = m.dot(&pixel);

            let pixel = pixel.mapv(RGB::apply_gamma).mapv(rescale_pixel);

            res.slice_mut(s![i, j, ..]).assign(&pixel);
        });
        Self::from_data(res)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, RGB>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        image.into_type_raw()
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, HSV>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, HSI>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, HSL>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, YCrCb>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, CIEXYZ>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, CIELAB>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic3>> for ImageBase<T, CIELUV>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic3>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic1>> for ImageBase<T, Gray>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic1>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Generic4>> for ImageBase<T, RGBA>
where
    T: Data,
{
    fn from(image: ImageBase<T, Generic4>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, RGB>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, RGB>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, HSV>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, HSV>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, HSI>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, HSI>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, HSL>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, HSL>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, YCrCb>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, YCrCb>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, CIEXYZ>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, CIEXYZ>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, CIELAB>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, CIELAB>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, CIELUV>> for ImageBase<T, Generic3>
where
    T: Data,
{
    fn from(image: ImageBase<T, CIELUV>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, RGBA>> for ImageBase<T, Generic4>
where
    T: Data,
{
    fn from(image: ImageBase<T, RGBA>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T> From<ImageBase<T, Gray>> for ImageBase<T, Generic1>
where
    T: Data,
{
    fn from(image: ImageBase<T, Gray>) -> Self {
        Self::from_data(image.data)
    }
}

impl<T, U> From<ImageBase<U, Generic5>> for Image<T, Generic4>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic5>) -> Self {
        let shape = (image.rows(), image.cols(), Generic4::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic5>> for Image<T, Generic3>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic5>) -> Self {
        let shape = (image.rows(), image.cols(), Generic3::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic5>> for Image<T, Generic2>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic5>) -> Self {
        let shape = (image.rows(), image.cols(), Generic2::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic5>> for Image<T, Generic1>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic5>) -> Self {
        let shape = (image.rows(), image.cols(), Generic1::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic4>> for Image<T, Generic3>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic4>) -> Self {
        let shape = (image.rows(), image.cols(), Generic3::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic4>> for Image<T, Generic2>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic4>) -> Self {
        let shape = (image.rows(), image.cols(), Generic2::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic4>> for Image<T, Generic1>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic4>) -> Self {
        let shape = (image.rows(), image.cols(), Generic1::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic3>> for Image<T, Generic2>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic3>) -> Self {
        let shape = (image.rows(), image.cols(), Generic2::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic3>> for Image<T, Generic1>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic3>) -> Self {
        let shape = (image.rows(), image.cols(), Generic1::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic2>> for Image<T, Generic1>
where
    U: Data<Elem = T>,
    T: Copy,
{
    fn from(image: ImageBase<U, Generic2>) -> Self {
        let shape = (image.rows(), image.cols(), Generic1::channels());
        let data = Array3::from_shape_fn(shape, |(i, j, k)| image.data[[i, j, k]]);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic1>> for Image<T, Generic5>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic1>) -> Self {
        let shape = (image.rows(), image.cols(), Generic5::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic1::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic2>> for Image<T, Generic5>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic2>) -> Self {
        let shape = (image.rows(), image.cols(), Generic5::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic2::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic3>> for Image<T, Generic5>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic3>) -> Self {
        let shape = (image.rows(), image.cols(), Generic5::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic3::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic4>> for Image<T, Generic5>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic4>) -> Self {
        let shape = (image.rows(), image.cols(), Generic5::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic4::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic1>> for Image<T, Generic4>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic1>) -> Self {
        let shape = (image.rows(), image.cols(), Generic4::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic1::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic2>> for Image<T, Generic4>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic2>) -> Self {
        let shape = (image.rows(), image.cols(), Generic4::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic2::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic3>> for Image<T, Generic4>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic3>) -> Self {
        let shape = (image.rows(), image.cols(), Generic4::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic3::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic1>> for Image<T, Generic3>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic1>) -> Self {
        let shape = (image.rows(), image.cols(), Generic3::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic1::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic2>> for Image<T, Generic3>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic2>) -> Self {
        let shape = (image.rows(), image.cols(), Generic3::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic2::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl<T, U> From<ImageBase<U, Generic1>> for Image<T, Generic2>
where
    U: Data<Elem = T>,
    T: Copy + Num,
{
    fn from(image: ImageBase<U, Generic1>) -> Self {
        let shape = (image.rows(), image.cols(), Generic2::channels());
        let mut data = Array3::zeros(shape);
        data.slice_mut(s![.., .., 0..Generic1::channels()])
            .assign(&image.data);
        Self::from_data(data)
    }
}

impl ColourModel for RGB {}
impl ColourModel for HSV {}
impl ColourModel for HSI {}
impl ColourModel for HSL {}
impl ColourModel for YCrCb {}
impl ColourModel for CIEXYZ {}
impl ColourModel for CIELAB {}
impl ColourModel for CIELUV {}

impl ColourModel for Gray {
    fn channels() -> usize {
        1
    }
}

impl ColourModel for Generic1 {
    fn channels() -> usize {
        1
    }
}
impl ColourModel for Generic2 {
    fn channels() -> usize {
        2
    }
}
impl ColourModel for Generic3 {
    fn channels() -> usize {
        3
    }
}
impl ColourModel for Generic4 {
    fn channels() -> usize {
        4
    }
}
impl ColourModel for Generic5 {
    fn channels() -> usize {
        5
    }
}

impl ColourModel for RGBA {
    fn channels() -> usize {
        4
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use ndarray::s;
    use ndarray_rand::RandomExt;
    use ndarray_stats::QuantileExt;
    use rand::distributions::Uniform;

    #[test]
    fn basic_rgb_hsv_check() {
        let mut i = Image::<u8, RGB>::new(1, 2);
        i.pixel_mut(0, 0).assign(&arr1(&[0, 0, 0]));
        i.pixel_mut(0, 1).assign(&arr1(&[255, 255, 255]));

        let hsv = Image::<u8, HSV>::from(i.clone());

        assert_eq!(hsv.pixel(0, 0)[[2]], 0);
        assert_eq!(hsv.pixel(0, 1)[[2]], 255);

        let rgb = Image::<u8, RGB>::from(hsv);
        assert_eq!(i, rgb);
    }

    #[test]
    fn gray_to_rgb_test() {
        let mut image = Image::<u8, Gray>::new(480, 640);
        let new_data = Array3::<u8>::random(image.data.dim(), Uniform::new(0, 255));
        image.data = new_data;

        let rgb = Image::<u8, RGB>::from(image.clone());
        let slice_2d = image.data.slice(s![.., .., 0]);

        assert_eq!(slice_2d, rgb.data.slice(s![.., .., 0]));
        assert_eq!(slice_2d, rgb.data.slice(s![.., .., 1]));
        assert_eq!(slice_2d, rgb.data.slice(s![.., .., 2]));
    }

    #[test]
    fn rgb_to_gray_basic() {
        // Check white, black, red, green, blue
        let data = vec![255, 255, 255, 0, 0, 0, 255, 0, 0, 0, 255, 0, 0, 0, 255];
        let image = Image::<u8, RGB>::from_shape_data(1, 5, data);

        let gray = Image::<u8, Gray>::from(image);

        // take standard 0.3 0.59 0.11 values and assume truncation
        let expected = vec![255, 0, 77, 150, 28];

        for (act, exp) in gray.data.iter().zip(expected.iter()) {
            let delta = (*act as i16 - *exp as i16).abs();
            assert!(delta < 2);
        }
    }

    #[test]
    fn basic_xyz_rgb_checks() {
        let mut image = Image::<f32, RGB>::new(100, 100);
        let new_data = Array3::<f32>::random(image.data.dim(), Uniform::new(0.0, 1.0));
        image.data = new_data;

        let xyz = Image::<f32, CIEXYZ>::from(image.clone());

        let rgb_restored = Image::<f32, RGB>::from(xyz);

        let mut delta = image.data - rgb_restored.data;
        delta.mapv_inplace(|x| x.abs());

        // 0.5% error in RGB -> XYZ -> RGB
        assert!(*delta.max().unwrap() * 100.0 < 0.5);
    }

    #[test]
    fn generic3_checks() {
        let mut image = Image::<f32, RGB>::new(100, 100);
        let new_data = Array3::<f32>::random(image.data.dim(), Uniform::new(0.0, 1.0));
        image.data = new_data;
        let gen = Image::<_, Generic3>::from(image.clone());
        // Normally don't check floats with equality but data shouldn't have
        // changed
        assert_eq!(gen.data, image.data);

        let hsv = Image::<_, HSV>::from(gen);
        assert_eq!(image.data, hsv.data);

        let gen = Image::<_, Generic3>::from(hsv);
        assert_eq!(image.data, gen.data);

        let hsi = Image::<_, HSI>::from(gen);
        assert_eq!(image.data, hsi.data);

        let gen = Image::<_, Generic3>::from(hsi);
        assert_eq!(image.data, gen.data);

        let hsl = Image::<_, HSL>::from(gen);
        assert_eq!(image.data, hsl.data);

        let gen = Image::<_, Generic3>::from(hsl);
        assert_eq!(image.data, gen.data);

        let ycrcb = Image::<_, YCrCb>::from(gen);
        assert_eq!(image.data, ycrcb.data);

        let gen = Image::<_, Generic3>::from(ycrcb);
        assert_eq!(image.data, gen.data);

        let ciexyz = Image::<_, CIEXYZ>::from(gen);
        assert_eq!(image.data, ciexyz.data);

        let gen = Image::<_, Generic3>::from(ciexyz);
        assert_eq!(image.data, gen.data);

        let cieluv = Image::<_, CIELUV>::from(gen);
        assert_eq!(image.data, cieluv.data);

        let gen = Image::<_, Generic3>::from(cieluv);
        assert_eq!(image.data, gen.data);

        let cielab = Image::<_, CIELAB>::from(gen);
        assert_eq!(image.data, cielab.data);
    }

    #[test]
    fn generic_model_expand() {
        let mut image = Image::<f32, Generic1>::new(100, 100);
        let new_data = Array3::<f32>::random(image.data.dim(), Uniform::new(0.0, 1.0));
        image.data = new_data;

        let large = Image::<_, Generic5>::from(image.clone());

        assert_eq!(large.channels(), Generic5::channels());

        assert_eq!(
            large.data.slice(s![.., .., 0]),
            image.data.slice(s![.., .., 0])
        );
        let zeros = Array2::<f32>::zeros((100, 100));
        for i in 1..Generic5::channels() {
            assert_eq!(large.data.slice(s![.., .., i]), zeros);
        }
    }
}