2 * Copyright (C) 2004 John Ellis
3 * Copyright (C) 2008 - 2016 The Geeqie Team
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
29 * These functions are intended to find images with similar color content. For
30 * example when an image was saved at different compression levels or dimensions
31 * (scaled down/up) the contents are similar, but these files do not match by file
32 * size, dimensions, or checksum.
34 * These functions create a 32 x 32 array for each color channel (red, green, blue).
35 * The array represents the average color of each corresponding part of the
36 * image. (imagine the image cut into 1024 rectangles, or a 32 x 32 grid.
37 * Each grid is then processed for the average color value, this is what
38 * is stored in the array)
40 * To compare two images, generate a ImageSimilarityData for each image, then
41 * pass them to the compare function. The return value is the percent match
42 * of the two images. (for this, simple comparisons are used, basically the return
43 * is an average of the corresponding array differences)
45 * for image_sim_compare(), the return is 0.0 to 1.0: \n
46 * 1.0 for exact matches (an image is compared to itself) \n
47 * 0.0 for exact opposite images (compare an all black to an all white image) \n
48 * generally only a match of > 0.85 are significant at all, and >.95 is useful to
49 * find images that have been re-saved to other formats, dimensions, or compression.
53 * The experimental (alternate) algorithm is only for testing of new techniques to
54 * improve the result, and hopes to reduce false positives.
57 static gboolean alternate_enabled = FALSE;
59 void image_sim_alternate_set(gboolean enable)
61 alternate_enabled = enable;
64 #pragma GCC diagnostic push
65 #pragma GCC diagnostic ignored "-Wunused-function"
66 gboolean image_sim_alternate_enabled_unused(void)
68 return alternate_enabled;
70 #pragma GCC diagnostic pop
72 ImageSimilarityData *image_sim_new()
74 auto sd = g_new0(ImageSimilarityData, 1);
79 void image_sim_free(ImageSimilarityData *sd)
84 static gint image_sim_channel_eq_sort_cb(gconstpointer a, gconstpointer b)
86 auto pa = static_cast<const gint *>(a);
87 auto pb = static_cast<const gint *>(b);
88 if (pa[1] < pb[1]) return -1;
89 if (pa[1] > pb[1]) return 1;
93 static void image_sim_channel_equal(guint8 *pix, gint len)
99 buf = g_new0(gint, len * 2);
102 for (i = 0; i < len; i++)
106 buf[p] = static_cast<gint>(pix[i]);
110 qsort(buf, len, sizeof(gint) * 2, image_sim_channel_eq_sort_cb);
113 for (i = 0; i < len; i++)
119 pix[n] = static_cast<guint8>(255 * i / len);
125 static void image_sim_channel_norm(guint8 *pix, gint len)
133 for (i = 0; i < len; i++)
135 if (pix[i] < l) l = pix[i];
136 if (pix[i] > h) h = pix[i];
140 scale = (delta != 0) ? 255.0 / static_cast<gdouble>(delta) : 1.0;
142 for (i = 0; i < len; i++)
144 pix[i] = static_cast<guint8>(static_cast<gdouble>(pix[i] - l) * scale);
149 * define these to enable various components of the experimental compare functions
151 * Convert the thumbprint to greyscale (ignore all color information when comparing)
152 * #define ALTERNATE_USES_GREYSCALE 1
154 * Take into account the difference in change from one pixel to the next
155 * #define ALTERNATE_INCLUDE_COMPARE_CHANGE 1
158 void image_sim_alternate_processing(ImageSimilarityData *sd)
160 #ifdef ALTERNATE_USES_GREYSCALE
164 if (!alternate_enabled) return;
166 image_sim_channel_norm(sd->avg_r, sizeof(sd->avg_r));
167 image_sim_channel_norm(sd->avg_g, sizeof(sd->avg_g));
168 image_sim_channel_norm(sd->avg_b, sizeof(sd->avg_b));
170 image_sim_channel_equal(sd->avg_r, sizeof(sd->avg_r));
171 image_sim_channel_equal(sd->avg_g, sizeof(sd->avg_g));
172 image_sim_channel_equal(sd->avg_b, sizeof(sd->avg_b));
174 #ifdef ALTERNATE_USES_GREYSCALE
175 for (i = 0; i < sizeof(sd->avg_r); i++)
179 n = (guint8)((gint)(sd->avg_r[i] + sd->avg_g[i] + sd->avg_b[i]) / 3);
180 sd->avg_r[i] = sd->avg_g[i] = sd->avg_b[i] = n;
185 gint mround(gdouble x)
188 gdouble fpart = x-ipart;
189 return (fpart < 0.5 ? ipart : ipart+1);
192 void image_sim_fill_data(ImageSimilarityData *sd, GdkPixbuf *pixbuf)
203 gint x_inc, y_inc, xy_inc;
207 gboolean x_small = FALSE; /* if less than 32 w or h, set TRUE */
208 gboolean y_small = FALSE;
209 if (!sd || !pixbuf) return;
211 w = gdk_pixbuf_get_width(pixbuf);
212 h = gdk_pixbuf_get_height(pixbuf);
213 rs = gdk_pixbuf_get_rowstride(pixbuf);
214 pix = gdk_pixbuf_get_pixels(pixbuf);
215 has_alpha = gdk_pixbuf_get_has_alpha(pixbuf);
217 p_step = has_alpha ? 4 : 3;
237 for (ys = 0; ys < 32; ys++)
239 if (y_small) j = static_cast<gdouble>(h) / 32 * ys;
240 else y_inc = mround(static_cast<gdouble>(h_left)/(32-ys));
244 for (xs = 0; xs < 32; xs++)
251 if (x_small) i = static_cast<gdouble>(w) / 32 * xs;
252 else x_inc = mround(static_cast<gdouble>(w_left)/(32-xs));
253 xy_inc = x_inc * y_inc;
255 xpos = pix + (i * p_step);
257 for (y = j; y < j + y_inc; y++)
260 for (x = i; x < i + x_inc; x++)
289 ImageSimilarityData *image_sim_new_from_pixbuf(GdkPixbuf *pixbuf)
291 ImageSimilarityData *sd;
293 sd = image_sim_new();
294 image_sim_fill_data(sd, pixbuf);
299 #ifdef ALTERNATE_INCLUDE_COMPARE_CHANGE
300 static gdouble alternate_image_sim_compare_fast(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min)
307 if (!a || !b || !a->filled || !b->filled) return 0.0;
313 for (j = 0; j < 1024; j += 32)
315 for (i = j; i < j + 32; i++)
320 cr = abs(a->avg_r[i] - b->avg_r[i]);
321 cg = abs(a->avg_g[i] - b->avg_g[i]);
322 cb = abs(a->avg_b[i] - b->avg_b[i]);
325 sim += cd + abs(cd - ld);
328 /* check for abort, if so return 0.0 */
329 if ((gdouble)sim / (255.0 * 1024.0 * 4.0) > min) return 0.0;
332 return (1.0 - ((gdouble)sim / (255.0 * 1024.0 * 4.0)) );
336 gdouble image_sim_compare_transfo(ImageSimilarityData *a, ImageSimilarityData *b, gchar transfo)
342 if (!a || !b || !a->filled || !b->filled) return 0.0;
346 if (transfo & 1) { i = &j2; j = &i2; } else { i = &i2; j = &j2; }
347 for (j1 = 0; j1 < 32; j1++)
349 if (transfo & 2) *j = 31-j1; else *j = j1;
350 for (i1 = 0; i1 < 32; i1++)
352 if (transfo & 4) *i = 31-i1; else *i = i1;
353 sim += abs(a->avg_r[i1*32+j1] - b->avg_r[i2*32+j2]);
354 sim += abs(a->avg_g[i1*32+j1] - b->avg_g[i2*32+j2]);
355 sim += abs(a->avg_b[i1*32+j1] - b->avg_b[i2*32+j2]);
359 return 1.0 - (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0));
362 gdouble image_sim_compare(ImageSimilarityData *a, ImageSimilarityData *b)
364 gint max_t = (options->rot_invariant_sim ? 8 : 1);
367 gdouble score, max_score = 0;
369 for(t = 0; t < max_t; t++)
371 score = image_sim_compare_transfo(a, b, t);
372 if (score > max_score) max_score = score;
379 4 rotations (0, 90, 180, 270) combined with two mirrors (0, H)
380 generate all possible isometric transformations
382 = change dir of x, change dir of y, exchange x and y = 2^3 = 8
384 gdouble image_sim_compare_fast_transfo(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min, gchar transfo)
390 #ifdef ALTERNATE_INCLUDE_COMPARE_CHANGE
391 if (alternate_enabled) return alternate_image_sim_compare_fast(a, b, min);
394 if (!a || !b || !a->filled || !b->filled) return 0.0;
399 if (transfo & 1) { i = &j2; j = &i2; } else { i = &i2; j = &j2; }
400 for (j1 = 0; j1 < 32; j1++)
402 if (transfo & 2) *j = 31-j1; else *j = j1;
403 for (i1 = 0; i1 < 32; i1++)
405 if (transfo & 4) *i = 31-i1; else *i = i1;
406 sim += abs(a->avg_r[i1*32+j1] - b->avg_r[i2*32+j2]);
407 sim += abs(a->avg_g[i1*32+j1] - b->avg_g[i2*32+j2]);
408 sim += abs(a->avg_b[i1*32+j1] - b->avg_b[i2*32+j2]);
410 /* check for abort, if so return 0.0 */
411 if (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0) > min) return 0.0;
414 return (1.0 - (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0)) );
417 /* this uses a cutoff point so that it can abort early when it gets to
418 * a point that can simply no longer make the cut-off point.
420 gdouble image_sim_compare_fast(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min)
422 gint max_t = (options->rot_invariant_sim ? 8 : 1);
425 gdouble score, max_score = 0;
427 for(t = 0; t < max_t; t++)
429 score = image_sim_compare_fast_transfo(a, b, min, t);
430 if (score > max_score) max_score = score;
434 /* vim: set shiftwidth=8 softtabstop=0 cindent cinoptions={1s: */