imageproc/
seam_carving.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
//! An implementation of [seam carving]. Currently in a pretty rough state.
//! See examples/seam_carving.rs for an example.
//!
//! [seam carving]: https://en.wikipedia.org/wiki/Seam_carving

use crate::definitions::{HasBlack, Image};
use crate::gradients::sobel_gradient_map;
use crate::map::{map_colors, WithChannel};
use image::{GrayImage, Luma, Pixel, Rgb};
use std::cmp::min;

/// An image seam connecting the bottom of an image to its top (in that order).
pub struct VerticalSeam(Vec<u32>);

/// Reduces the width of an image using seam carving.
///
/// Warning: this is very slow! It implements the algorithm from
/// https://inst.eecs.berkeley.edu/~cs194-26/fa16/hw/proj4-seamcarving/imret.pdf, with some
/// extra unnecessary allocations thrown in. Rather than attempting to optimise the implementation
/// of this inherently slow algorithm, the planned next step is to switch to the algorithm from
/// https://users.cs.cf.ac.uk/Paul.Rosin/resources/papers/seam-carving-ChinaF.pdf.
pub fn shrink_width<P>(image: &Image<P>, target_width: u32) -> Image<P>
// TODO: this is pretty silly! We should just be able to express that we want a pixel which is a slice of integral values
where
    P: Pixel<Subpixel = u8> + WithChannel<u16> + WithChannel<i16>,
    <P as WithChannel<u16>>::Pixel: HasBlack,
{
    assert!(
        target_width <= image.width(),
        "target_width must be <= input image width"
    );

    let iterations = image.width() - target_width;
    let mut result = image.clone();

    for _ in 0..iterations {
        let seam = find_vertical_seam(&result);
        result = remove_vertical_seam(&result, &seam);
    }

    result
}

/// Computes an 8-connected path from the bottom of the image to the top whose sum of
/// gradient magnitudes is minimal.
pub fn find_vertical_seam<P>(image: &Image<P>) -> VerticalSeam
where
    P: Pixel<Subpixel = u8> + WithChannel<u16> + WithChannel<i16>,
    <P as WithChannel<u16>>::Pixel: HasBlack,
{
    let (width, height) = image.dimensions();
    assert!(
        image.width() >= 2,
        "Cannot find seams if image width is < 2"
    );

    let mut gradients = sobel_gradient_map(image, |p| {
        let gradient_sum: u16 = p.channels().iter().sum();
        let gradient_mean: u16 = gradient_sum / P::CHANNEL_COUNT as u16;
        Luma([gradient_mean as u32])
    });

    // Find the least energy path through the gradient image.
    for y in 1..height {
        for x in 0..width {
            set_path_energy(&mut gradients, x, y);
        }
    }

    // Retrace our steps to find the vertical seam.
    let mut min_x = 0;
    let mut min_energy = gradients.get_pixel(0, height - 1)[0];

    for x in 1..width {
        let c = gradients.get_pixel(x, height - 1)[0];
        if c < min_energy {
            min_x = x;
            min_energy = c;
        }
    }

    let mut seam = Vec::with_capacity(height as usize);

    seam.push(min_x);

    let mut last_x = min_x;

    for y in (1..height).rev() {
        let above = gradients.get_pixel(last_x, y - 1)[0];
        if last_x > 0 {
            let left = gradients.get_pixel(last_x - 1, y - 1)[0];
            if left < above {
                min_x = last_x - 1;
                min_energy = left;
            }
        }
        if last_x < width - 1 {
            let right = gradients.get_pixel(last_x + 1, y - 1)[0];
            if right < min_energy {
                min_x = last_x + 1;
                min_energy = right;
            }
        }

        last_x = min_x;
        seam.push(min_x);
    }

    VerticalSeam(seam)
}

/// Assumes that the previous rows have all been processed.
fn set_path_energy(path_energies: &mut Image<Luma<u32>>, x: u32, y: u32) {
    let above = path_energies.get_pixel(x, y - 1)[0];
    let mut min_energy = above;

    if x > 0 {
        let above_left = path_energies.get_pixel(x - 1, y - 1)[0];
        min_energy = min(above, above_left);
    }
    if x < path_energies.width() - 1 {
        let above_right = path_energies.get_pixel(x + 1, y - 1)[0];
        min_energy = min(min_energy, above_right);
    }

    let current = path_energies.get_pixel(x, y)[0];
    path_energies.put_pixel(x, y, Luma([min_energy + current]));
}

/// Returns the result of removing `seam` from `image`.
// This should just mutate an image in place. The problem is that we don't have a
// way of talking about views of ImageBuffer without devolving into supporting
// arbitrary GenericImages. And a lot of other functions don't support those because
// it would make them a lot slower.
pub fn remove_vertical_seam<P>(image: &Image<P>, seam: &VerticalSeam) -> Image<P>
where
    P: Pixel,
{
    assert!(
        seam.0.len() as u32 == image.height(),
        "seam length does not match image height"
    );

    let (width, height) = image.dimensions();
    let mut out = Image::new(width - 1, height);

    for y in 0..height {
        let x_seam = seam.0[(height - y - 1) as usize];
        for x in 0..x_seam {
            out.put_pixel(x, y, *image.get_pixel(x, y));
        }
        for x in (x_seam + 1)..width {
            out.put_pixel(x - 1, y, *image.get_pixel(x, y));
        }
    }

    out
}

/// Draws a series of `seams` on `image` in red. Assumes that the provided seams were
/// removed in the given order from the input image.
pub fn draw_vertical_seams(image: &GrayImage, seams: &[VerticalSeam]) -> Image<Rgb<u8>> {
    let height = image.height();

    let mut offsets = vec![vec![]; height as usize];
    let mut out = map_colors(image, |p| p.to_rgb());

    for seam in seams {
        for (y, x) in (0..height).rev().zip(&seam.0) {
            let mut x_original = *x;
            for o in &offsets[y as usize] {
                if *o < *x {
                    x_original += 1;
                }
            }
            out.put_pixel(x_original, y, Rgb([255, 0, 0]));
            offsets[y as usize].push(x_original);
        }
    }

    out
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::utils::gray_bench_image;
    use test::{black_box, Bencher};

    macro_rules! bench_shrink_width {
        ($name:ident, side: $s:expr, shrink_by: $m:expr) => {
            #[bench]
            fn $name(b: &mut Bencher) {
                let image = gray_bench_image($s, $s);
                b.iter(|| {
                    let filtered = shrink_width(&image, $s - $m);
                    black_box(filtered);
                })
            }
        };
    }

    bench_shrink_width!(bench_shrink_width_s100_r1, side: 100, shrink_by: 1);
    bench_shrink_width!(bench_shrink_width_s100_r4, side: 100, shrink_by: 4);
    bench_shrink_width!(bench_shrink_width_s100_r8, side: 100, shrink_by: 8);
}