利用k-means聚类算法识别图片主色调

由于使用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 代码如下复制代码 $start = microtime(TRUE); main(); function main($img = ‘colors_files/T1OX3eXldXXXcqfYM._111424.jpg’) { list($width, $height, $mime_code) = getimagesize($img); $im = null; $point = array(); switch ($mime_code) { # jpg case 2: $im =imagecreatefromjpeg($img); break; # png case 3: default: exit(‘擦 ，什么图像？解析不了啊’); } $new_width = 100; $new_height = 100; $pixel = imagecreatetruecolor($new_width, $new_height); imagecopyresampled($pixel, $im, 0, 0, 0, 0, $new_width, $new_height, $width, $height); run_time(); $i = $new_width; while ($i–) { # reset高度 $k = $new_height; while ($k–) { $rgb = ImageColorAt($im, $i, $k); array_push($point, array(‘r’=>($rgb >> 16) & 0xFF, ‘g’=>($rgb >> 8) & 0xFF, ‘b’=>$rgb & 0xFF)); } } imagedestroy($im); imagedestroy($pixel); run_time(); $color = kmeans($point); run_time(); foreach ($color as $key => $value)   { echo ‘’ . RGBToHex($value[0]) . ‘’; } } function run_time() { global $start; echo ‘消耗:’, microtime(TRUE) – $start; } function kmeans($point=array(), $k=3, $min_diff=1) { global $ii; $point_len = count($point); $clusters = array(); $cache = array(); for ($i=0; $i { $cache[$i] = $i*$i; } # 随机生成k值 $i = $k; $index = 0; while ($i–) { $index = mt_rand(1,$point_len-100); array_push($clusters, array($point[$index], array($point[$index]))); } run_time(); $point_list = array(); $run_num = 0; while (TRUE) { foreach ($point as $value) { $smallest_distance = 10000000; # 求出距离最小的点 # index用于保存point最靠近的k值 $index = 0; $i = $k; while ($i–) { $distance = 0; foreach ($value as $key => $p1) {   if ($p1 > $clusters[$i][0][$key]) { $distance += $cache[$p1 - $clusters[$i][0][$key]]; } else { $distance += $cache[$clusters[$i][0][$key] – $p1]; } } $ii++; if ($distance { $smallest_distance = $distance; $index = $i; } } $point_list[$index][] = $value; } $diff = 0; # 1个1个迭代k值 $i = $k; while ($i–) { $old = $clusters[$i]; # 移到到队列中心 $center = calculateCenter($point_list[$i], 3); # 形成新的k值集合队列 $new_cluster = array($center, $point_list[$i]); $clusters[$i] = $new_cluster; # 计算新的k值与队列所在点的位置 $diff = euclidean($old[0], $center); } # 判断是否已足够聚合 if ($diff { break; >} } echo ‘—>’.$ii; return $clusters; } # 计算2点距离 $ii = 0; function euclidean($p1, $p2) { $s = 0; foreach ($p1 as $key => $value) { $temp = ($value – $p2[$key]); $s += $temp*$temp; } return sqrt($s); } # 移动k值到所有点的中心 function calculateCenter($point_list, $attr_num) { $vals = array(); $point_num = 0; $keys = array_keys($point_list[0]); foreach($keys as $value) { $vals[$value] = 0; } foreach ($point_list as $arr) { $point_num++; foreach ($arr as $key => $value) { $vals[$key] += $value; } } foreach ($keys as $index) { $vals[$index] = $vals[$index] / $point_num; } return $vals; } function RGBToHex($r, $g=”, $b=”) { if (is_array($r)) { $b = $r['b']; $g = $r['g']; $r = $r['r']; } $hex = “#”; $hex.= str_pad(dechex($r), 2, ’0′, STR_PAD_LEFT); $hex.= str_pad(dechex($g), 2, ’0′, STR_PAD_LEFT); $hex.= str_pad(dechex($b), 2, ’0′, STR_PAD_LEFT); return $hex; } ?>