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
|
use super::Pixel;
pub struct NeuQuantQuantiser {
weights: [[f64; 3]; 256],
freq: [f64; 256],
bias: [f64; 256],
factor: usize,
}
const SPECIAL_NODES: usize = 2;
impl NeuQuantQuantiser {
pub fn new(factor: usize) -> Self {
let mut weights = [[0.0; 3]; 256];
if SPECIAL_NODES > 1 {
weights[1] = [255.0; 3]; // for white
}
for i in SPECIAL_NODES..256 {
let w = 255.0 * ((i - SPECIAL_NODES) as f64) / ((256 - SPECIAL_NODES) as f64);
weights[i] = [w, w, w];
}
Self {
weights,
freq: [1.0 / 256.0; 256],
bias: [0.0; 256],
factor,
}
}
fn update_node(&mut self, idx: usize, clr: &[f64; 3], alpha: f64) {
self.weights[idx][0] -= alpha * (self.weights[idx][0] - clr[0]);
self.weights[idx][1] -= alpha * (self.weights[idx][1] - clr[1]);
self.weights[idx][2] -= alpha * (self.weights[idx][2] - clr[2]);
}
fn update_neighbours(&mut self, idx: usize, clr: &[f64; 3], alpha: f64, radius: usize) {
let low = idx.saturating_sub(radius).max(SPECIAL_NODES - 1);
let high = (idx + radius).min(self.weights.len() - 1);
let mut idx0 = idx + 1;
let mut idx1 = idx - 1;
let mut range = 0;
let sqradius = (radius * radius) as f64;
while (idx0 < high) || (idx1 > low) {
let sqrng = f64::from(range * range);
let a = alpha * (sqradius - sqrng) / sqradius;
range += 1;
if idx0 < high {
self.update_node(idx0, clr, a);
idx0 += 1;
}
if idx1 > low {
self.update_node(idx1, clr, a);
idx1 -= 1;
}
}
}
fn find_node(&mut self, clr: &[f64; 3]) -> usize {
for i in 0..SPECIAL_NODES {
if &self.weights[i] == clr {
return i;
}
}
let mut bestdist = std::f64::MAX;
let mut distidx = 0;
let mut bestbias = std::f64::MAX;
let mut biasidx = 0;
for i in SPECIAL_NODES..256 {
let dist = (self.weights[i][0] - clr[0]) * (self.weights[i][0] - clr[0])
+ (self.weights[i][1] - clr[1]) * (self.weights[i][1] - clr[1])
+ (self.weights[i][2] - clr[2]) * (self.weights[i][2] - clr[2]);
if bestdist > dist {
bestdist = dist;
distidx = i;
}
let biasdiff = dist - self.bias[i];
if bestbias > biasdiff {
bestbias = biasdiff;
biasidx = i;
}
self.freq[i] -= self.freq[i] / 1024.0;
self.bias[i] += self.freq[i];
}
self.freq[distidx] += 1.0 / 1024.0;
self.bias[distidx] -= 1.0;
biasidx
}
pub fn learn(&mut self, src: &[Pixel]) {
let mut bias_radius = (256 / 8) << 6;
let alphadec = (30 + (self.factor - 1) / 3) as f64;
let initial_alpha = f64::from(1 << 10);
let npixels = src.len();
let mut radius = bias_radius >> 6;
if radius == 1 { radius = 0 };
let samples = npixels / self.factor;
let delta = samples / 100;
let mut alpha = initial_alpha;
let mut pos = 0;
const PRIMES: [usize; 4] = [ 499, 491, 487, 503 ];
let mut step = PRIMES[3];
for prime in PRIMES.iter().rev() {
if npixels % *prime != 0 {
step = *prime;
}
}
for i in 0..samples {
let clr = [f64::from(src[pos].r), f64::from(src[pos].g), f64::from(src[pos].b)];
let idx = self.find_node(&clr);
if idx >= SPECIAL_NODES {
let new_alpha = alphadec / initial_alpha;
self.update_node(idx, &clr, new_alpha);
if radius > 0 {
self.update_neighbours(idx, &clr, new_alpha, radius);
}
}
pos = (pos + step) % npixels;
if (i + 1) % delta == 0 {
alpha -= alpha / alphadec;
bias_radius -= bias_radius / 30;
radius = bias_radius >> 6;
if radius == 1 { radius = 0 };
}
}
}
pub fn make_pal(&self, pal: &mut [[u8; 3]; 256]) {
for (pal, node) in pal.iter_mut().zip(self.weights.iter()) {
pal[0] = (node[0] + 0.5).max(0.0).min(255.0) as u8;
pal[1] = (node[1] + 0.5).max(0.0).min(255.0) as u8;
pal[2] = (node[2] + 0.5).max(0.0).min(255.0) as u8;
}
}
}
|
