本例仅以本人浅薄理解，妄想自制文字识别程序，实际在识别部分未有完善。
<?php
class readChar{
private $imgSize; //图片尺寸
private $imgGd2; //图像转GD2
private $Index=array(); //颜色索引(key即为颜色索引)
private $bigColor; //二维图像颜色值(存储索引)
function __construct($imgPath){
$this->imgSize=getimagesize($imgPath);
$this->imgSize['size']=$this->imgSize[0]*$this->imgSize[1];
$this->imgGd2=imagecreatefromstring(file_get_contents($imgPath));
if (imageistruecolor($this->imgGd2)) {
imagetruecolortopalette($this->imgGd2, false, 256);//真彩图片转换为调色板
}
$this->setGray();
}
function __destruct(){
imagedestroy($this->imgGd2);
}
private function showImg(){
foreach($this->Index as $k=>$v){
imagecolorset($this->imgGd2,$k,$v,$v,$v);
}
header('Content-type: image/jpg');
imagejpeg($this->imgGd2);
exit;
}
private function setGray(){
/*
灰度化
RGB均值/RGB单值/最大/最小/人性化：0.3R+0.59G+0.11B
bug:若灰度值相等的两个颜色，刚好是主要颜色 则会识别不出来
*/
for($i=ImageColorstotal($this->imgGd2)-1;$i>=0;$i--){
$rgb=ImageColorsForIndex($this->imgGd2,$i);
$this->Index[$i]=(int)(($rgb['red']+$rgb['green']+$rgb['blue'])/3); //imagecolorset改变索引颜色
}
$this->bigColor=array();
$pro=array(); //各灰度值占比
for($x=0;$x<$this->imgSize[0];$x++){
$this->bigColor[$x]=array();
for($y=0;$y<$this->imgSize[1];$y++){
$Index=ImageColorAt($this->imgGd2, $x, $y);
$this->bigColor[$x][$y]=$Index;
$pro[$this->Index[$Index]]=@$pro[$this->Index[$Index]]+1;
}
}
array_walk($pro,function(&$v){$v=$v/$this->imgSize['size'];});
$this->setTwo($pro); }
private function setTwo($pro){
/*
二值化 T很重要
以T为阈值,低于T的为白否则为黑
双峰法
迭代法：
OSTU（大津法）:不懂
前景和背景的分割阈值记作T，前景像素点数占比为ω0，平均灰度μ0；背景像素点数占比例ω1，平均灰度为μ1。
总平均灰度记为μ
类间方差记假设图像的背景较暗，并且图像的大小为M×N，灰度值小于阈值T的像素数为N0，大于阈值T的像素数为N1
则有：
ω0=N0/ M×N (1)
ω1=N1/ M×N (2)
N0+N1=M×N (3)
ω0+ω1=1 (4)
μ=ω0*μ0+ω1*μ1 (5)
g=ω0(μ0-μ)^2+ω1(μ1-μ)^2 (6)
将式(5)代入式(6),得到等价公式: g=ω0ω1(μ0-μ1)^2
类间方差g最大时的阈值T,即为所求
P分位法:需已知目标占图像的比例，以不同灰度值进行分割若比例≈P 则T为该灰度值
*/
$T=127;
$g_max=0;
for ($i=0;$i<256;$i++){
$w0 = $w1 = $u0_temp = $u1_temp = $u0 = $u1 = $g_tmp = 0;
for ($j=0;$j<256;$j++){
if ($j <= $i){ //背景部分
$w0 += @$pro[$j];
$u0_temp += $j * @$pro[$j];
}else{ //前景部分
$w1 += @$pro[$j];
$u1_temp += $j * @$pro[$j];
}
}
$u0 = $w0==0?0:$u0_temp / $w0;
$u1 = $w1==0?0:$u1_temp / $w1;
$g_tmp =$w0 *$w1* pow(($u0 - $u1), 2);//类间方差 g=w0*w1*(u0-u1)^2
if ($g_tmp > $g_max){
$g_max = $g_tmp;
$T = $i;
}
}
for($x=0;$x<$this->imgSize[0];$x++){
for($y=0;$y<$this->imgSize[1];$y++){
$index = $this->bigColor[$x][$y];
if($this->Index[$index]<=$T){
$this->Index[$index]=0;
}else{
$this->Index[$index]=255;
}
}
}
$this->avgFilter();
}
private function avgFilter(){
/*
代码不实现
均值滤波器、自适应维纳滤波器、中值滤波器、形态学噪声滤除器、小波去噪
滤波前对于图片边界：不处理/填充0 or 255/填充临近灰度值
*/
return $this->getChar(); }
private function getChar(){
/*
拆字
*/
$pointTotal=array(); //Y轴统计
for($x=0;$x<$this->imgSize[0];$x++){
for($y=0;$y<$this->imgSize[1];$y++){
@$pointTotal[$y]+=$this->Index[$this->bigColor[$x][$y]]>0?0:1;
}
}
$chars=array(); //Y轴划线
$prev = $pointTotal[0];
$tmpLine=array();
foreach($pointTotal as $k=>$v){
if($v==0 && $prev!=0){
//imageline ($this->imgGd2,0,$k,$this->imgSize[0]-1,$k,0);//划线 对程序无用
$tmpLine[]=$k;
}elseif($v!=0 && $prev==0){
//imageline ($this->imgGd2,0,$k-1,$this->imgSize[0]-1,$k,0);//划线 对程序无用
$tmpLine[]=$k-1;
}
$prev=$v;
if(count($tmpLine)==2){
$chars[]=$tmpLine;
$tmpLine=array();
}
}
if(!$chars){
//imageline ($this->imgGd2,0,0,$this->imgSize[0]-1,0,0);//划线 对程序无用
//imageline ($this->imgGd2,0,$this->imgSize[1]-1,$this->imgSize[0]-1,$this->imgSize[1]-1,0);//划线 对程序无用
$chars []=array(0,$this->imgSize[1]-1);
}
foreach($chars as $line=>$ypoint){
$pointTotal=array();//每行的X轴统计
for($x=0;$x<$this->imgSize[0];$x++){
$pointTotal[$x]=0;
for($y=$ypoint[0];$y<=$ypoint[1];$y++){
$pointTotal[$x]+=$this->Index[$this->bigColor[$x][$y]]>0?0:1;
}
}
$xLine=array();
$tmpLine=array();//每行X轴划线
$prev = $pointTotal[0];
foreach($pointTotal as $k=>$v){
if($v==0 && $prev!=0){
//imageline ($this->imgGd2,$k,$ypoint[0],$k,$ypoint[1],0);//划线 对程序无用
$tmpLine[]=$k-1;
}
if($v!=0 && $prev==0){
//imageline ($this->imgGd2,$k-1,$ypoint[0],$k-1,$ypoint[1],0);//划线 对程序无用
$tmpLine[]=$k;
}
if(count($tmpLine)==2){
$xLine[]=$tmpLine;
$tmpLine=array();
}
$prev=$v;
}
foreach($xLine as $k=>$v){
$v['xcode']=$v['ycode']=array();
for($x=$v[0];$x<=$v[1];$x++){
for($y=$ypoint[0];$y<=$ypoint[1];$y++){
$gry = $this->Index[$this->bigColor[$x][$y]]>0?0:1;
@$v['xcode'][$x-$v[0]] +=$gry;
@$v['ycode'][$y-$ypoint[0]] +=$gry;
}
}
$xLine[$k]=$v;
}
$chars[$line]['xline']=$xLine;
}
$this->bigColor=null;
foreach($chars as $v){
foreach($v['xline'] as $vv){
$this->tranChar($vv['xcode'],$vv['ycode']);
}
}
}
private function tranChar($myX,$myY){
/*
识别文字
本例用到的php自带函数 similar_text
通过把每个字x和y轴做映射，然后和模板做相似度匹配（模板图为50x50所以需将映射做压缩处理）
*/
$tplx='0,0,0,0,0,0,0,0,12,22,30,34,23,16,13,11,10,8,8,8,8,8,8,6,6,6,6,8,8,7,8,9,10,10,12,14,20,34,30,26,16,0,0,0,0,0,0,0,0,0';
$tply='9,14,17,15,11,10,10,8,8,8,9,8,8,7,8,8,7,8,7,8,8,8,8,8,8,8,8,8,8,8,8,7,8,7,8,8,7,8,8,8,8,9,8,9,10,12,15,17,13,9';
$diff=count($myX)-count($myY);
$middle = (int)(abs($diff)/2);
if($diff<0){
$minMy=&$myX;
}else{
$minMy=&$myY;
}
for($i=0;$i<abs($diff);$i++){
if($i<$middle){
array_unshift($minMy,0);
continue;
}
array_push($minMy,0);
}
$ratio = 50/count($myX);
$newX=array();
$newY=array();
foreach($myX as $k=>$v){
$key = min(ceil($k*$ratio),49);
is_array(@$newX[$key]) || $newX[$key]=array();
is_array(@$newY[$key]) || $newY[$key]=array();
$newX[$key][]=$myX[$k];
$newY[$key][]=$myY[$k];
}
array_walk($newY,function(&$v){$v=round(array_sum($v)/count($v));});
array_walk($newX,function(&$v){$v=round(array_sum($v)/count($v));}); $sx=similar_text(implode(',',$newX),$tplx);
$sy=similar_text(implode(',',$newY),$tply);
echo 'X:'.$sx.'/'.strlen($tplx).'='.($sx/strlen($tplx));
echo "<br>";
echo 'Y:'.$sy.'/'.strlen($tply).'='.($sy/strlen($tply));
exit;
}
}
new readChar("imgurl.jpg");
附上模板图片：
