nihav-core/src/scale/palette/elbg.rs

   1 use super::Pixel;
   2
   3 struct RNG {
   4     seed: u16,
   5 }
   6
   7 impl RNG {
   8     fn new() -> Self { Self { seed: 0x1234 } }
   9     fn next(&mut self) -> u8 {
  10         if (self.seed & 0x8000) != 0 {
  11             self.seed = ((self.seed & 0x7FFF) * 2) ^ 0x1B2B;
  12         } else {
  13             self.seed <<= 1;
  14         }
  15         self.seed as u8
  16     }
  17 }
  18
  19 #[derive(Default,Clone,Copy,PartialEq,Debug)]
  20 struct Entry {
  21     pix:        Pixel,
  22     count:      u64,
  23 }
  24
  25 struct Cluster {
  26     centroid:   Pixel,
  27     dist:       u64,
  28     count:      u64,
  29     sum_r:      u64,
  30     sum_g:      u64,
  31     sum_b:      u64,
  32 }
  33
  34 impl Cluster {
  35     fn new(centroid: Pixel) -> Self {
  36         Self {
  37             centroid,
  38             dist:       0,
  39             count:      0,
  40             sum_r:      0,
  41             sum_g:      0,
  42             sum_b:      0,
  43         }
  44     }
  45     fn reset(&mut self) {
  46         self.count = 0;
  47         self.sum_r = 0;
  48         self.sum_g = 0;
  49         self.sum_b = 0;
  50         self.dist  = 0;
  51     }
  52     fn add_pixel(&mut self, entry: &Entry) {
  53         self.sum_r += u64::from(entry.pix.r) * entry.count;
  54         self.sum_g += u64::from(entry.pix.g) * entry.count;
  55         self.sum_b += u64::from(entry.pix.b) * entry.count;
  56         self.count += entry.count;
  57     }
  58     fn add_dist(&mut self, entry: &Entry) {
  59         self.dist += u64::from(self.centroid.dist(entry.pix)) * entry.count;
  60     }
  61     fn calc_centroid(&mut self) {
  62         if self.count != 0 {
  63             self.centroid.r = ((self.sum_r + self.count / 2) / self.count) as u8;
  64             self.centroid.g = ((self.sum_g + self.count / 2) / self.count) as u8;
  65             self.centroid.b = ((self.sum_b + self.count / 2) / self.count) as u8;
  66         }
  67     }
  68     fn calc_dist(&mut self) {
  69     }
  70 }
  71
  72 pub struct ELBG {
  73     clusters:   Vec<Cluster>,
  74 }
  75
  76 impl ELBG {
  77     #[allow(dead_code)]
  78     pub fn new(initial_pal: &[[u8; 3]; 256]) -> Self {
  79         let mut clusters = Vec::with_capacity(256);
  80         for i in 0..256 {
  81             let pix = Pixel { r: initial_pal[i][0], g: initial_pal[i][1], b: initial_pal[i][2] };
  82             let cluster = Cluster::new(pix);
  83             clusters.push(cluster);
  84         }
  85         Self {
  86             clusters,
  87         }
  88     }
  89     #[allow(dead_code)]
  90     pub fn new_random() -> Self {
  91         let mut rng = RNG::new();
  92         let mut clusters = Vec::with_capacity(256);
  93         for _ in 0..256 {
  94             let pix = Pixel { r: rng.next(), g: rng.next(), b: rng.next() };
  95             let cluster = Cluster::new(pix);
  96             clusters.push(cluster);
  97         }
  98         Self {
  99             clusters,
 100         }
 101     }
 102     fn sort<F>(arr: &mut [Pixel], idx: F) where F: Fn(&Pixel) -> u8 {
 103         let mut dst = vec![Pixel::default(); arr.len()];
 104         let mut counts = [0; 256];
 105         for pix in arr.iter() {
 106             counts[idx(pix) as usize] += 1;
 107         }
 108         let mut last = counts[0];
 109         counts[0] = 0;
 110         for count in counts.iter_mut().skip(1) {
 111             let plast = last;
 112             last += *count;
 113             *count = plast;
 114         }
 115         for pix in arr.iter() {
 116             let bucket = idx(pix) as usize;
 117             dst[counts[bucket]] = *pix;
 118             counts[bucket] += 1;
 119         }
 120         arr.copy_from_slice(dst.as_slice());
 121     }
 122     fn new_split(old_index: usize, entries: &[Entry], indices: &[usize]) -> Option<(Pixel, Pixel)> {
 123         let mut max = Pixel { r:   0, g:   0, b:   0 };
 124         let mut min = Pixel { r: 255, g: 255, b: 255 };
 125         let mut found = false;
 126         for (entry, idx) in entries.iter().zip(indices) {
 127             if *idx == old_index {
 128                 max = max.max(entry.pix);
 129                 min = min.min(entry.pix);
 130                 found = true;
 131             }
 132         }
 133         if !found {
 134             return None;
 135         }
 136         let dr = max.r - min.r;
 137         let dg = max.g - min.g;
 138         let db = max.b - min.b;
 139         let cent0 = Pixel { r: min.r + dr / 3, g: min.g + dg / 3, b: min.b + db / 3 };
 140         let cent1 = Pixel { r: max.r - dr / 3, g: max.g - dg / 3, b: max.b - db / 3 };
 141         Some((cent0, cent1))
 142     }
 143     fn old_centre(&self, old_index1: usize, old_index2: usize, entries: &[Entry], indices: &[usize]) -> Pixel {
 144         let mut max = Pixel { r:   0, g:   0, b:   0 };
 145         let mut min = Pixel { r: 255, g: 255, b: 255 };
 146         let mut found = false;
 147         for (entry, idx) in entries.iter().zip(indices) {
 148             if *idx == old_index1 || *idx == old_index2 {
 149                 max = max.max(entry.pix);
 150                 min = min.min(entry.pix);
 151                 found = true;
 152             }
 153         }
 154         if !found {
 155             max = self.clusters[old_index1].centroid.max(self.clusters[old_index2].centroid);
 156             min = self.clusters[old_index1].centroid.min(self.clusters[old_index2].centroid);
 157         }
 158         let dr = max.r - min.r;
 159         let dg = max.g - min.g;
 160         let db = max.b - min.b;
 161         Pixel { r: min.r + dr / 2, g: min.g + dg / 2, b: min.b + db / 2 }
 162     }
 163     fn estimate_old(old_idx0: usize, old_idx1: usize, c: Pixel, entries: &[Entry], indices: &[usize]) -> u64 {
 164         let mut clu = Cluster::new(c);
 165         let mut count = 0;
 166         for (entry, idx) in entries.iter().zip(indices) {
 167             if *idx == old_idx0 || *idx == old_idx1 {
 168                 clu.add_dist(entry);
 169                 count += entry.count;
 170             }
 171         }
 172         clu.count = count;
 173         clu.calc_dist();
 174         clu.dist
 175     }
 176     fn estimate_new(c0: Pixel, c1: Pixel, old_idx: usize, entries: &[Entry], indices: &[usize]) -> u64 {
 177         let mut clu0 = Cluster::new(c0);
 178         let mut clu1 = Cluster::new(c1);
 179         let mut count0 = 0;
 180         let mut count1 = 0;
 181         for (entry, idx) in entries.iter().zip(indices) {
 182             if *idx == old_idx {
 183                 if c0.dist(entry.pix) < c1.dist(entry.pix) {
 184                     clu0.add_dist(entry);
 185                     count0 += entry.count;
 186                 } else {
 187                     clu1.add_dist(entry);
 188                     count1 += entry.count;
 189                 }
 190             }
 191         }
 192         clu0.count = count0;
 193         clu1.count = count1;
 194         clu0.calc_dist();
 195         clu1.calc_dist();
 196         clu0.dist + clu1.dist
 197     }
 198     #[allow(clippy::cognitive_complexity)]
 199     pub fn quantise(&mut self, src: &[Pixel], dst: &mut [[u8; 3]; 256]) {
 200         if src.len() < 3 {
 201             return;
 202         }
 203         let mut old_cb: [Pixel; 256] = [Pixel::default(); 256];
 204         let mut prev_dist = std::u64::MAX;
 205         let mut dist = std::u64::MAX / 2;
 206         let mut indices = Vec::with_capacity(src.len());
 207         let mut pixels = Vec::with_capacity(src.len());
 208         pixels.extend_from_slice(src);
 209         Self::sort(pixels.as_mut_slice(), |pix| pix.r);
 210         Self::sort(pixels.as_mut_slice(), |pix| pix.g);
 211         Self::sort(pixels.as_mut_slice(), |pix| pix.b);
 212         let mut entries = Vec::with_capacity(pixels.len() / 2);
 213         let mut lastval = pixels[0];
 214         let mut run = 1;
 215         for pix in pixels.iter().skip(1) {
 216             if &lastval == pix {
 217                 run += 1;
 218             } else {
 219                 entries.push(Entry { pix: lastval, count: run });
 220                 lastval = *pix;
 221                 run = 1;
 222             }
 223         }
 224         entries.push(Entry { pix: lastval, count: run });
 225         drop(pixels);
 226
 227         let mut low_u:  Vec<usize> = Vec::with_capacity(256);
 228         let mut high_u: Vec<usize> = Vec::with_capacity(256);
 229         let mut rng = RNG::new();
 230         let mut iterations = 0usize;
 231         let mut do_elbg_step = true;
 232         while (iterations < 10) && (dist < prev_dist - prev_dist / 100) {
 233             prev_dist = dist;
 234             for i in 0..256 {
 235                 old_cb[i] = self.clusters[i].centroid;
 236                 self.clusters[i].reset();
 237             }
 238             // put pixels into the nearest clusters
 239             indices.clear();
 240             for entry in entries.iter() {
 241                 let mut bestidx = 0;
 242                 let mut bestdist = std::u32::MAX;
 243                 for (i, cluster) in self.clusters.iter().enumerate() {
 244                     let dist = entry.pix.dist(cluster.centroid);
 245                     if bestdist > dist {
 246                         bestdist = dist;
 247                         bestidx = i;
 248                         if dist == 0 {
 249                             break;
 250                         }
 251                     }
 252                 }
 253                 indices.push(bestidx);
 254                 self.clusters[bestidx].add_pixel(entry);
 255             }
 256             // calculate params
 257             for cluster in self.clusters.iter_mut() {
 258                 cluster.calc_centroid();
 259             }
 260             dist = 0;
 261             for (idx, entry) in indices.iter().zip(entries.iter()) {
 262                 self.clusters[*idx].add_dist(entry);
 263             }
 264             for cluster in self.clusters.iter_mut() {
 265                 cluster.calc_dist();
 266                 dist += cluster.dist;
 267             }
 268
 269             let dmean = dist / 256;
 270             low_u.clear();
 271             high_u.clear();
 272             let mut used = [false; 256];
 273             for (i, cluster) in self.clusters.iter().enumerate() {
 274                 if cluster.dist < dmean {
 275                     low_u.push(i);
 276                 } else if cluster.dist > dmean * 2 {
 277                     high_u.push(i);
 278                     used[i] = true;
 279                 }
 280             }
 281
 282             if do_elbg_step {
 283                 do_elbg_step = false;
 284                 for low_idx in low_u.iter() {
 285                     if high_u.is_empty() {
 286                         break;
 287                     }
 288                     let high_idx_idx = (rng.next() as usize) % high_u.len();
 289                     let high_idx = high_u[high_idx_idx];
 290                     let mut closest_idx = *low_idx;
 291                     let mut closest_dist = std::u32::MAX;
 292                     let low_centr = self.clusters[*low_idx].centroid;
 293                     for i in 0..256 {//low_u.iter() {
 294                         if i == *low_idx || used[i] {
 295                             continue;
 296                         }
 297                         let dist = self.clusters[i].centroid.dist(low_centr);
 298                         if closest_dist > dist {
 299                             closest_dist = dist;
 300                             closest_idx  = i;
 301                         }
 302                     }
 303                     if closest_idx == *low_idx {
 304                         continue;
 305                     }
 306                     let old_dist = self.clusters[*low_idx].dist + self.clusters[closest_idx].dist + self.clusters[high_idx].dist;
 307                     let old_centr = self.old_centre(*low_idx, closest_idx, entries.as_slice(), indices.as_slice());
 308                     let ret = Self::new_split(high_idx, entries.as_slice(), indices.as_slice());
 309                     if ret.is_none() {
 310                         continue;
 311                     }
 312                     let (centr0, centr1) = ret.unwrap();
 313                     let dist_o = if old_dist > self.clusters[high_idx].dist {
 314                             Self::estimate_old(*low_idx, closest_idx, old_centr, entries.as_slice(), indices.as_slice())
 315                         } else { 0 };
 316                     let dist_n = Self::estimate_new(centr0, centr1, high_idx, entries.as_slice(), indices.as_slice());
 317                     if dist_o + dist_n < old_dist {
 318                         self.clusters[*low_idx   ].centroid = old_centr;
 319                         self.clusters[closest_idx].centroid = centr0;
 320                         self.clusters[high_idx   ].centroid = centr1;
 321                         used[*low_idx]    = true;
 322                         used[closest_idx] = true;
 323                         used[high_idx]    = true;
 324                         high_u.remove(high_idx_idx);
 325                         do_elbg_step = true;
 326                     }
 327                 }
 328             }
 329             iterations += 1;
 330         }
 331         if dist < prev_dist {
 332             for i in 0..256 {
 333                 old_cb[i] = self.clusters[i].centroid;
 334             }
 335         }
 336         for i in 0..256 {
 337             dst[i][0] = old_cb[i].r;
 338             dst[i][1] = old_cb[i].g;
 339             dst[i][2] = old_cb[i].b;
 340         }
 341     }
 342 }