vftdan
/
cc-stuff
forked from hkc/cc-stuff
1
0
Fork 0
Code Pull Requests Activity

Compare commits

base: vftdan:c5af524dac5839e6db6bc2a92576d7f3b43e42c4
Branches Tags
vftdan:master
vftdan:palette-functions
vftdan:optimized-lookup
vftdan:palette-k-means
vftdan:optimized8x11test
vftdan:vftdan
hkc:master
hkc:augment/dev
...
compare: vftdan:5f95f895d245200c33b92ba8978b4ae30dec4657
Branches Tags
vftdan:palette-functions
vftdan:optimized-lookup
vftdan:palette-k-means
vftdan:optimized8x11test
vftdan:vftdan
vftdan:master
hkc:master
hkc:augment/dev

2 Commits
c5af524dac ... 5f95f895d2

Author SHA1 Message Date
Vftdan 5f95f895d2 Try to avoid having unused palette items when the source image only uses a limited gamut region 
Warp each empty cluster's centroid onto the closest dataset item
 2024-10-03 10:20:58 +02:00
Vftdan c91f4d79bd Allocate the correct amount of intermediate centroid data 2024-10-03 10:16:57 +02:00
1 changed files with 19 additions and 4 deletions
Show Stats Expand all files Collapse all files

23
img2cpi.c
View File

@ -95,6 +95,8 @@ struct k_means_state {
float r, g, b;
} sums;
size_t count;
union color closest_present_item;
float closest_present_distance;
} *centroid_intermediate;
size_t item_count;
};
@ -675,13 +677,20 @@ struct palette *custom_palette_from(const struct palette *orig) {
struct k_means_state k_means_init(const struct image *image, struct palette *starting_palette) {
size_t item_count = image->w * image->h;
uint8_t cluster_count = starting_palette->count;
struct k_means_state state = {
.items = image,
.clusters = starting_palette,
.predicted_cluster = calloc(image->w, image->h),
.centroid_intermediate = calloc(item_count, sizeof(struct k_means_centroid_intermediate)),
.centroid_intermediate = calloc(cluster_count, sizeof(struct k_means_centroid_intermediate)),
.item_count = item_count,
};
if (state.centroid_intermediate) {
for (size_t i = 0; i < cluster_count; i++) {
state.centroid_intermediate[i].closest_present_item = starting_palette->colors[i];
state.centroid_intermediate[i].closest_present_distance = 1e20;
}
}
return state;
}
@ -696,11 +705,16 @@ bool k_means_iteration(struct k_means_state *state) {
int closest_cluster = 0;
float closest_distance = 1e20;
for (int cluster = 0; cluster < state->clusters->count; cluster++) {
float dist = get_color_difference(state->clusters->colors[cluster], state->items->pixels[i]);
union color item = state->items->pixels[i];
float dist = get_color_difference(state->clusters->colors[cluster], item);
if (dist <= closest_distance) {
closest_distance = dist;
closest_cluster = cluster;
}
if (dist <= state->centroid_intermediate[cluster].closest_present_distance) {
state->centroid_intermediate[cluster].closest_present_item = item;
state->centroid_intermediate[cluster].closest_present_distance = dist;
}
}
if (!changed) {
changed = state->predicted_cluster[i] != closest_cluster;
@ -730,9 +744,10 @@ bool k_means_iteration(struct k_means_state *state) {
state->clusters->colors[i] = centroid;
} else {
// No pixels are closest to this color
// TODO: wiggle the centroid?
// Warp the centroid onto the closest item
state->clusters->colors[i] = intermediate.closest_present_item;
}
state->centroid_intermediate[i] = (struct k_means_centroid_intermediate) { .sums = {0, 0, 0}, .count = 0 };
state->centroid_intermediate[i] = (struct k_means_centroid_intermediate) { .sums = {0, 0, 0}, .count = 0, .closest_present_item = state->clusters->colors[i], .closest_present_distance = 1e20 };
}
return changed;