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

use std::ops::Mul;

use image::Rgba32FImage;
use nalgebra::Matrix3;
use nalgebra::{ArrayStorage, Matrix1x2, Matrix2, Matrix2x1, Matrix3x1};

pub fn warp_into(input: &Rgba32FImage, matrix: Matrix3<f32>, output: &mut Rgba32FImage) {
    let inverse = matrix.try_inverse().unwrap();

    let in_width = input.width();
    let in_height = input.height();

    let out_width = output.width();
    let out_height = output.height();

    for out_row in 0..out_width {
        for out_col in 0..out_height {
            let point = Matrix3x1::<f32>::new(out_row as f32, out_col as f32, 1f32);

            let in_pixel = inverse * point;

            let in_row = in_pixel.x as i32;
            let in_col = in_pixel.y as i32;

            if (0 <= in_row)
                && (in_row < in_width as i32)
                && (0 <= in_col)
                && (in_col < in_height as i32)
            {
                let px = input.get_pixel(in_row as _, in_col as _);

                output[(out_row, out_col)] = *px;
            }
        }
    }
}

/// The Kabsch-Umeyama algorithm is a method for finding the optimal translation, rotation,
/// and scaling that aligns two sets of points with minimum root-mean-square deviation (RMSD).
pub fn umeyama<const R: usize>(src: &[(f32, f32); R], dst: &[(f32, f32); R]) -> Matrix3<f32> {
    let src_x_sum: f32 = src.iter().map(|v| v.0).sum();
    let src_x_mean = src_x_sum / (R as f32);

    let src_y_sum: f32 = src.iter().map(|v| v.1).sum();
    let src_y_mean = src_y_sum / (R as f32);

    let dst_x_sum: f32 = dst.iter().map(|v| v.0).sum();
    let dst_x_mean = dst_x_sum / (R as f32);

    let dst_y_sum: f32 = dst.iter().map(|v| v.1).sum();
    let dst_y_mean = dst_y_sum / (R as f32);

    let src_demean_s = ArrayStorage(src.map(|v| [v.0 - src_x_mean, v.1 - src_y_mean]));
    let dst_demean_s = ArrayStorage(dst.map(|v| [v.0 - dst_x_mean, v.1 - dst_y_mean]));

    let src_demean = nalgebra::Matrix::from_array_storage(src_demean_s);
    let dst_demean = nalgebra::Matrix::from_array_storage(dst_demean_s);

    let a = std::ops::Mul::mul(dst_demean, &src_demean.transpose()) / (R as f32);
    let svd = nalgebra::Matrix::svd(a, true, true);

    let determinant = a.determinant();

    let mut d = [1f32; 2];

    if determinant < 0.0f32 {
        d[2 - 1] = -1.0f32;
    }

    let mut t = Matrix2::<f32>::identity();
    let s = svd.singular_values;
    let u = svd.u.unwrap();
    let v = svd.v_t.unwrap();

    let rank = a.rank(0.00001f32);

    if rank == 0 {
        panic!("Matrix rank is 0.");
    } else if rank == 2 - 1 {
        //
        if u.determinant() * v.determinant() > 0.0 {
            u.mul_to(&v, &mut t);
        } else {
            let s = d[2 - 1];
            d[2 - 1] = -1f32;
            let dg = Matrix2::<f32>::new(d[0], 0f32, 0f32, d[1]);

            let udg = u.mul(&dg);
            udg.mul_to(&v, &mut t);
            d[2 - 1] = s;
        }
    } else {
        let dg = Matrix2::<f32>::new(d[0], 0f32, 0f32, d[1]);
        let udg = u.mul(&dg);
        udg.mul_to(&v, &mut t);
    }

    let ddd = Matrix1x2::new(d[0], d[1]);
    let d_x_s = ddd.mul(s);

    let var0 = src_demean.remove_row(0).variance();
    let var1 = src_demean.remove_row(1).variance();

    let varsum = var0 + var1;

    let scale = d_x_s.get((0, 0)).unwrap() / varsum;

    let dst_mean = Matrix2x1::<f32>::new(dst_x_mean, dst_y_mean);
    let src_mean = Matrix2x1::<f32>::new(src_x_mean, src_y_mean);
    let t_x_srcmean = t.mul(&src_mean);

    let xxx = scale * t_x_srcmean;
    let yyy = dst_mean - xxx;

    let m13 = *yyy.get(0).unwrap();
    let m23 = *yyy.get(1).unwrap();

    let m00x22 = t * scale;

    let m11 = m00x22.m11;
    let m21 = m00x22.m21;
    let m12 = m00x22.m12;
    let m22 = m00x22.m22;

    let tt = Matrix3::<f32>::new(m11, m12, m13, m21, m22, m23, 0f32, 0f32, 1f32);

    return tt;
}

#[cfg(test)]
mod tests {
    use super::*;
    const SRC: [(f32, f32); 5] = [
        (491.7426, 321.8467),
        (541.7967, 332.23264),
        (507.47015, 366.4012),
        (485.72678, 369.63214),
        (533.7206, 378.40567),
    ];

    const DST: [(f32, f32); 5] = [
        (38.2946f32, 51.6963f32),
        (73.5318f32, 51.5014f32),
        (56.0252f32, 71.7366f32),
        (41.5493f32, 92.3655f32),
        (70.7299f32, 92.2041f32),
    ];

    #[test]
    fn estimate() {
        let result = umeyama(&SRC, &DST);

        const R: nalgebra::Matrix<
            f32,
            nalgebra::Const<3>,
            nalgebra::Const<3>,
            ArrayStorage<f32, 3, 3>,
        > = Matrix3::<f32>::new(
            0.7000762f32,
            0.12366913f32,
            -346.2191f32,
            -0.12366913f32,
            0.7000762f32,
            -112.38885f32,
            0f32,
            0f32,
            1f32,
        );

        assert_eq!(result, R);
    }
}