由于使用php来写图片主色调识别功能太麻烦了,所以我给大家介绍利用利用k-means聚类算法识别图片主色调方法.

识别图片主色调这个,网上貌似有几种方法,不过最准确,最优雅的解决方案还是利用聚类算法来做.

直接上代码,不过,我测试结果表示,用PHP来做,效率不佳,PHP不适合做这种大规模运算,用nodejs做效率可以高出100倍左右,代码如下:

1. <?php
2. $start = microtime(TRUE);
3. main();
4. function main($img = ‘colors_files/T1OX3eXldXXXcqfYM._111424.jpg’)
5. {
6. list($width, $height, $mime_code) = getimagesize($img);
7. //开源代码phpfensi.com
8. $im = null;
9. $point = array();
10. switch ($mime_code)
11. {
12. # jpg
13. case 2:
14. $im =imagecreatefromjpeg($img);
15. break;
16. # png
17. case 3:
18. default:
19. exit(‘擦 ，什么图像？解析不了啊’);
20. }
21. $new_width = 100;
22. $new_height = 100;
23. $pixel = imagecreatetruecolor($new_width, $new_height);
24. imagecopyresampled($pixel, $im, 0, 0, 0, 0, $new_width, $new_height, $width, $height);
25. run_time();
26. $i = $new_width;
27. while ($i–)
28. {
29. # reset高度
30. $k = $new_height;
31. while ($k–)
32. {
33. $rgb = ImageColorAt($im, $i, $k);
34. array_push($point, array(‘r’=>($rgb >> 16) & 0xFF, ‘g’=>($rgb >> 8) & 0xFF, ‘b’=>$rgb & 0xFF));
35. }
36. }
37. imagedestroy($im);
38. imagedestroy($pixel);
39. run_time();
40. $color = kmeans($point);
41. run_time();
42. foreach ($color as $key => $value)
43. &nb
44. sp; {
45. echo ‘<br><span style=“background-color:’ . RGBToHex($value[0]) . ‘” >’ . RGBToHex($value[0]) . ‘</span>’;
46. }
47. }
48. function run_time()
49. {
50. global $start;
51. echo ‘<br/>消耗:’, microtime(TRUE) – $start;
52. }
53. function kmeans($point=array(), $k=3, $min_diff=1)
54. {
55. global $ii;
56. $point_len = count($point);
57. $clusters = array();
58. $cache = array();
59. for ($i=0; $i < 256; $i++)
60. {
61. $cache[$i] = $i*$i;
62. }
63. # 随机生成k值
64. $i = $k;
65. $index = 0;
66. while ($i–)
67. {
68. $index = mt_rand(1,$point_len-100);
69. array_push($clusters, array($point[$index], array($point[$index])));
70. }
71. run_time();
72. $point_list = array();
73. $run_num = 0;
74. while (TRUE)
75. {
76. foreach ($point as $value)
77. {
78. $smallest_distance = 10000000;
79. # 求出距离最小的点
80. # index用于保存point最靠近的k值
81. $index = 0;
82. $i = $k;
83. while ($i–)
84. {
85. $distance = 0;
86. foreach ($value as $key => $p1)
87. {
88. &n
89. bsp; if ($p1 > $clusters[$i][0][$key])
90. {
91. $distance += $cache[$p1 - $clusters[$i][0][$key]];
92. }
93. else
94. {
95. $distance += $cache[$clusters[$i][0][$key] – $p1];
96. }
97. }
98. $ii++;
99. if ($distance < $smallest_distance)
100. {
101. $smallest_distance = $distance;
102. $index = $i;
103. }
104. }
105. $point_list[$index][] = $value;
106. }
107. $diff = 0;
108. # 1个1个迭代k值
109. $i = $k;
110. while ($i–)
111. {
112. $old = $clusters[$i];
113. # 移到到队列中心
114. $center = calculateCenter($point_list[$i], 3);
115. # 形成新的k值集合队列
116. $new_cluster = array($center, $point_list[$i]);
117. $clusters[$i] = $new_cluster;
118. # 计算新的k值与队列所在点的位置
119. $diff = euclidean($old[0], $center);
120. }
121. # 判断是否已足够聚合
122. if ($diff < $min_diff)
123. {
124. break;
125. >
126. }
127. }
128. echo ‘—>’.$ii;
129. return $clusters;
130. }
131. # 计算2点距离
132. $ii = 0;
133. function euclidean($p1, $p2)
134. {
135. $s = 0;
136. foreach ($p1 as $key => $value)
137. {
138. $temp = ($value – $p2[$key]);
139. $s += $temp*$temp;
140. }
141. return sqrt($s);
142. }
143. # 移动k值到所有点的中心
144. function calculateCenter($point_list, $attr_num) {
145. $vals = array();
146. $point_num = 0;
147. $keys = array_keys($point_list[0]);
148. foreach($keys as $value)
149. {
150. $vals[$value] = 0;
151. }
152. foreach ($point_list as $arr)
153. {
154. $point_num++;
155. foreach ($arr as $key => $value)
156. {
157. $vals[$key] += $value;
158. }
159. }
160. foreach ($keys as $index)
161. {
162. $vals[$index] = $vals[$index] / $point_num;
163. }
164. return $vals;
165. }
166. function RGBToHex($r, $g=”, $b=”)
167. {
168. if (is_array($r))
169. {
170. $b = $r['b'];
171. $g = $r['g'];
172. $r = $r['r'];
173. }
174. $hex = “#”;
175. $hex.= str_pad(dechex($r), 2, ’0′, STR_PAD_LEFT);
176. $hex.= str_pad(dechex($g), 2, ’0′, STR_PAD_LEFT);
177. $hex.= str_pad(dechex($b), 2, ’0′, STR_PAD_LEFT);
178. return $hex;
179. }
180. ?>