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.
52 ImageSimilarityData *image_sim_new()
54 auto sd = g_new0(ImageSimilarityData, 1);
59 void image_sim_free(ImageSimilarityData *sd)
64 static gint image_sim_channel_eq_sort_cb(gconstpointer a, gconstpointer b)
66 auto pa = static_cast<const gint *>(a);
67 auto pb = static_cast<const gint *>(b);
68 if (pa[1] < pb[1]) return -1;
69 if (pa[1] > pb[1]) return 1;
73 static void image_sim_channel_equal(guint8 *pix, gint len)
79 buf = g_new0(gint, len * 2);
82 for (i = 0; i < len; i++)
86 buf[p] = static_cast<gint>(pix[i]);
90 qsort(buf, len, sizeof(gint) * 2, image_sim_channel_eq_sort_cb);
93 for (i = 0; i < len; i++)
99 pix[n] = static_cast<guint8>(255 * i / len);
105 static void image_sim_channel_norm(guint8 *pix, gint len)
113 for (i = 0; i < len; i++)
115 if (pix[i] < l) l = pix[i];
116 if (pix[i] > h) h = pix[i];
120 scale = (delta != 0) ? 255.0 / static_cast<gdouble>(delta) : 1.0;
122 for (i = 0; i < len; i++)
124 pix[i] = static_cast<guint8>(static_cast<gdouble>(pix[i] - l) * scale);
129 * The Alternate algorithm is only for testing of new techniques to
130 * improve the result, and hopes to reduce false positives.
132 void image_sim_alternate_processing(ImageSimilarityData *sd)
136 if (!options->alternate_similarity_algorithm.enabled)
141 image_sim_channel_norm(sd->avg_r, sizeof(sd->avg_r));
142 image_sim_channel_norm(sd->avg_g, sizeof(sd->avg_g));
143 image_sim_channel_norm(sd->avg_b, sizeof(sd->avg_b));
145 image_sim_channel_equal(sd->avg_r, sizeof(sd->avg_r));
146 image_sim_channel_equal(sd->avg_g, sizeof(sd->avg_g));
147 image_sim_channel_equal(sd->avg_b, sizeof(sd->avg_b));
149 if (options->alternate_similarity_algorithm.grayscale)
151 for (i = 0; i < (gint)sizeof(sd->avg_r); i++)
155 n = (guint8)((gint)(sd->avg_r[i] + sd->avg_g[i] + sd->avg_b[i]) / 3);
156 sd->avg_r[i] = sd->avg_g[i] = sd->avg_b[i] = n;
161 gint mround(gdouble x)
164 gdouble fpart = x-ipart;
165 return (fpart < 0.5 ? ipart : ipart+1);
168 void image_sim_fill_data(ImageSimilarityData *sd, GdkPixbuf *pixbuf)
179 gint x_inc, y_inc, xy_inc;
183 gboolean x_small = FALSE; /* if less than 32 w or h, set TRUE */
184 gboolean y_small = FALSE;
185 if (!sd || !pixbuf) return;
187 w = gdk_pixbuf_get_width(pixbuf);
188 h = gdk_pixbuf_get_height(pixbuf);
189 rs = gdk_pixbuf_get_rowstride(pixbuf);
190 pix = gdk_pixbuf_get_pixels(pixbuf);
191 has_alpha = gdk_pixbuf_get_has_alpha(pixbuf);
193 p_step = has_alpha ? 4 : 3;
213 for (ys = 0; ys < 32; ys++)
215 if (y_small) j = static_cast<gdouble>(h) / 32 * ys;
216 else y_inc = mround(static_cast<gdouble>(h_left)/(32-ys));
220 for (xs = 0; xs < 32; xs++)
227 if (x_small) i = static_cast<gdouble>(w) / 32 * xs;
228 else x_inc = mround(static_cast<gdouble>(w_left)/(32-xs));
229 xy_inc = x_inc * y_inc;
231 xpos = pix + (i * p_step);
233 for (y = j; y < j + y_inc; y++)
236 for (x = i; x < i + x_inc; x++)
265 ImageSimilarityData *image_sim_new_from_pixbuf(GdkPixbuf *pixbuf)
267 ImageSimilarityData *sd;
269 sd = image_sim_new();
270 image_sim_fill_data(sd, pixbuf);
275 static gdouble alternate_image_sim_compare_fast(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min)
282 if (!a || !b || !a->filled || !b->filled) return 0.0;
288 for (j = 0; j < 1024; j += 32)
290 for (i = j; i < j + 32; i++)
295 cr = abs(a->avg_r[i] - b->avg_r[i]);
296 cg = abs(a->avg_g[i] - b->avg_g[i]);
297 cb = abs(a->avg_b[i] - b->avg_b[i]);
300 sim += cd + abs(cd - ld);
303 /* check for abort, if so return 0.0 */
304 if ((gdouble)sim / (255.0 * 1024.0 * 4.0) > min) return 0.0;
307 return (1.0 - ((gdouble)sim / (255.0 * 1024.0 * 4.0)) );
310 gdouble image_sim_compare_transfo(ImageSimilarityData *a, ImageSimilarityData *b, gchar transfo)
316 if (!a || !b || !a->filled || !b->filled) return 0.0;
320 if (transfo & 1) { i = &j2; j = &i2; } else { i = &i2; j = &j2; }
321 for (j1 = 0; j1 < 32; j1++)
323 if (transfo & 2) *j = 31-j1; else *j = j1;
324 for (i1 = 0; i1 < 32; i1++)
326 if (transfo & 4) *i = 31-i1; else *i = i1;
327 sim += abs(a->avg_r[i1*32+j1] - b->avg_r[i2*32+j2]);
328 sim += abs(a->avg_g[i1*32+j1] - b->avg_g[i2*32+j2]);
329 sim += abs(a->avg_b[i1*32+j1] - b->avg_b[i2*32+j2]);
333 return 1.0 - (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0));
336 gdouble image_sim_compare(ImageSimilarityData *a, ImageSimilarityData *b)
338 gint max_t = (options->rot_invariant_sim ? 8 : 1);
341 gdouble score, max_score = 0;
343 for(t = 0; t < max_t; t++)
345 score = image_sim_compare_transfo(a, b, t);
346 if (score > max_score) max_score = score;
353 4 rotations (0, 90, 180, 270) combined with two mirrors (0, H)
354 generate all possible isometric transformations
356 = change dir of x, change dir of y, exchange x and y = 2^3 = 8
358 gdouble image_sim_compare_fast_transfo(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min, gchar transfo)
364 if (options->alternate_similarity_algorithm.enabled)
366 return alternate_image_sim_compare_fast(a, b, min);
369 if (!a || !b || !a->filled || !b->filled) return 0.0;
374 if (transfo & 1) { i = &j2; j = &i2; } else { i = &i2; j = &j2; }
375 for (j1 = 0; j1 < 32; j1++)
377 if (transfo & 2) *j = 31-j1; else *j = j1;
378 for (i1 = 0; i1 < 32; i1++)
380 if (transfo & 4) *i = 31-i1; else *i = i1;
381 sim += abs(a->avg_r[i1*32+j1] - b->avg_r[i2*32+j2]);
382 sim += abs(a->avg_g[i1*32+j1] - b->avg_g[i2*32+j2]);
383 sim += abs(a->avg_b[i1*32+j1] - b->avg_b[i2*32+j2]);
385 /* check for abort, if so return 0.0 */
386 if (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0) > min) return 0.0;
389 return (1.0 - (static_cast<gdouble>(sim) / (255.0 * 1024.0 * 3.0)) );
392 /* this uses a cutoff point so that it can abort early when it gets to
393 * a point that can simply no longer make the cut-off point.
395 gdouble image_sim_compare_fast(ImageSimilarityData *a, ImageSimilarityData *b, gdouble min)
397 gint max_t = (options->rot_invariant_sim ? 8 : 1);
400 gdouble score, max_score = 0;
402 for(t = 0; t < max_t; t++)
404 score = image_sim_compare_fast_transfo(a, b, min, t);
405 if (score > max_score) max_score = score;
409 /* vim: set shiftwidth=8 softtabstop=0 cindent cinoptions={1s: */