Face detection in PHP

March 27, 2012
Posted by: Admin
Category: PHP

Face Detection in PHP (without OpenCV).

We have been looking for ways to detect faces in photos with PHP. Nowadays, face detection is built in many consumer products (camera obviously, but also Google and iPhoto), and seems to be a pretty common job. So we expected to find many solutions for doing it with PHP. Surprisingly, the only one we could find is OpenCV, an opensource lib that was originally developed by Intel. OpenCV seems to perform well but you need to be able to install it on your server. In our case, we wanted to have a pure PHP solution, so it can work with most hosts.

Learning about face detection

So, we started to think about implementing it our self. We read a articles, scientific papers, etc. The website http://www.facedetection.com/ is a great resource by the way. From this short research, we learnt that one of the most popular solutions is to use Viola Jones training with a Haar classifier. Very informative, but tedious to implement.

Always look at what others are doing

Then we looked for existing implementations in other languages. Let’s say Flash and Javascript. With Canvas, Javascript developers will certainly look at what flash developers do. There seem to be a few libs for face detection in AS3. They even work in real time. Pretty cool. We kept searching and finally found a canvas+javascript implementation of face detection at http://blog.kpicturebooth.com/?p=8. The code looked fairly compact and simple. Shouldn’t be hard to port to PHP.

Once the code converted to PHP, here’s the result:

<?php
class FaceDetector
{
    protected $detection_data;
    protected $canvas;
    protected $face;
    private $reduced_canvas;

    public function __construct($detection_file = 'detection.dat')
    {
        if (is_file($detection_file)) {
            $this->detection_data = unserialize(file_get_contents($detection_file));
        } else {
            throw new Exception("Couldn't load detection data");
        }
    }

    public function faceDetect($file)
    {
        if (is_resource($file)) {
            $this->canvas = $file;
        } elseif (is_file($file)) {
            $this->canvas = imagecreatefromjpeg($file);
        } else {
            throw new Exception("Can not load $file");
        }
        $im_width = imagesx($this->canvas);
        $im_height = imagesy($this->canvas);

        //Resample before detection?
        $ratio = 0;
        $diff_width = 320 - $im_width;
        $diff_height = 240 - $im_height;
        if ($diff_width > $diff_height) {
            $ratio = $im_width / 320;
        } else {
            $ratio = $im_height / 240;
        }

        if ($ratio != 0) {
            $this->reduced_canvas = imagecreatetruecolor($im_width / $ratio, $im_height / $ratio);
            imagecopyresampled(
                $this->reduced_canvas,
                $this->canvas,
                0,
                0,
                0,
                0,
                $im_width / $ratio,
                $im_height / $ratio,
                $im_width,
                $im_height
            );
            $stats = $this->getImgStats($this->reduced_canvas);
            $this->face = $this->doDetectGreedyBigToSmall(
                $stats['ii'],
                $stats['ii2'],
                $stats['width'],
                $stats['height']
            );
            if ($this->face['w'] > 0) {
                $this->face['x'] *= $ratio;
                $this->face['y'] *= $ratio;
                $this->face['w'] *= $ratio;
            }
        } else {
            $stats = $this->getImgStats($this->canvas);
            $this->face = $this->doDetectGreedyBigToSmall(
                $stats['ii'],
                $stats['ii2'],
                $stats['width'],
                $stats['height']
            );
        }
        return ($this->face['w'] > 0);
    }

    public function toJpeg()
    {
        $color = imagecolorallocate($this->canvas, 255, 0, 0); //red
        imagerectangle(
            $this->canvas,
            $this->face['x'],
            $this->face['y'],
            $this->face['x']+$this->face['w'],
            $this->face['y']+ $this->face['w'],
            $color
        );
        header('Content-type: image/jpeg');
        imagejpeg($this->canvas);
    }

    public function toJson()
    {
        return json_encode($this->face);
    }
    public function getFace()
    {
        return $this->face;
    }

    protected function getImgStats($canvas)
    {
        $image_width = imagesx($canvas);
        $image_height = imagesy($canvas);
        $iis =  $this->computeII($canvas, $image_width, $image_height);
        return array(
            'width' => $image_width,
            'height' => $image_height,
            'ii' => $iis['ii'],
            'ii2' => $iis['ii2']
        );
    }

    protected function computeII($canvas, $image_width, $image_height)
    {
        $ii_w = $image_width+1;
        $ii_h = $image_height+1;
        $ii = array();
        $ii2 = array();

        for ($i=0; $i<$ii_w; $i++) {
            $ii[$i] = 0;
            $ii2[$i] = 0;
        }
        for ($i=1; $i<$ii_h-1; $i++) {
            $ii[$i*$ii_w] = 0;
            $ii2[$i*$ii_w] = 0;
            $rowsum = 0;
            $rowsum2 = 0;
            for ($j=1; $j<$ii_w-1; $j++) {
                $rgb = ImageColorAt($canvas, $j, $i);
                $red = ($rgb >> 16) & 0xFF;
                $green = ($rgb >> 8) & 0xFF;
                $blue = $rgb & 0xFF;
                $grey = (0.2989*$red + 0.587*$green + 0.114*$blue)>>0;  // this is what matlab uses
                $rowsum += $grey;
                $rowsum2 += $grey*$grey;
                $ii_above = ($i-1)*$ii_w + $j;
                $ii_this = $i*$ii_w + $j;
                $ii[$ii_this] = $ii[$ii_above] + $rowsum;
                $ii2[$ii_this] = $ii2[$ii_above] + $rowsum2;
            }
        }
        return array('ii'=>$ii, 'ii2' => $ii2);
    }

    protected function doDetectGreedyBigToSmall($ii, $ii2, $width, $height)
    {
        $s_w = $width/20.0;
        $s_h = $height/20.0;
        $start_scale = $s_h < $s_w ? $s_h : $s_w;
        $scale_update = 1 / 1.2;
        for ($scale = $start_scale; $scale > 1; $scale *= $scale_update) {
            $w = (20*$scale) >> 0;
            $endx = $width - $w - 1;
            $endy = $height - $w - 1;
            $step = max($scale, 2) >> 0;
            $inv_area = 1 / ($w*$w);
            for ($y = 0; $y < $endy; $y += $step) {
                for ($x = 0; $x < $endx; $x += $step) {
                    $passed = $this->detectOnSubImage($x, $y, $scale, $ii, $ii2, $w, $width+1, $inv_area);
                    if ($passed) {
                        return array('x'=>$x, 'y'=>$y, 'w'=>$w);
                    }
                } // end x
            } // end y
        }  // end scale
        return null;
    }

    protected function detectOnSubImage($x, $y, $scale, $ii, $ii2, $w, $iiw, $inv_area)
    {
        $mean  = ($ii[($y+$w)*$iiw + $x + $w] + $ii[$y*$iiw+$x] - $ii[($y+$w)*$iiw+$x] - $ii[$y*$iiw+$x+$w])*$inv_area;

        $vnorm = ($ii2[($y+$w)*$iiw + $x + $w]
                  + $ii2[$y*$iiw+$x]
                  - $ii2[($y+$w)*$iiw+$x]
                  - $ii2[$y*$iiw+$x+$w])*$inv_area - ($mean*$mean);

        $vnorm = $vnorm > 1 ? sqrt($vnorm) : 1;
        $passed = true;
        for ($i_stage = 0; $i_stage < count($this->detection_data); $i_stage++) {
            $stage = $this->detection_data[$i_stage];
            $trees = $stage[0];
            $stage_thresh = $stage[1];
            $stage_sum = 0;

            for ($i_tree = 0; $i_tree < count($trees); $i_tree++) {
                $tree = $trees[$i_tree];
                $current_node = $tree[0];
                $tree_sum = 0;
                while ($current_node != null) {
                    $vals = $current_node[0];
                    $node_thresh = $vals[0];
                    $leftval = $vals[1];
                    $rightval = $vals[2];
                    $leftidx = $vals[3];
                    $rightidx = $vals[4];
                    $rects = $current_node[1];
                    $rect_sum = 0;
                    for ($i_rect = 0; $i_rect < count($rects); $i_rect++) {
                        $s = $scale;
                        $rect = $rects[$i_rect];
                        $rx = ($rect[0]*$s+$x)>>0;
                        $ry = ($rect[1]*$s+$y)>>0;
                        $rw = ($rect[2]*$s)>>0;
                        $rh = ($rect[3]*$s)>>0;
                        $wt = $rect[4];

                        $r_sum = ($ii[($ry+$rh)*$iiw + $rx + $rw]
                                  + $ii[$ry*$iiw+$rx]
                                  - $ii[($ry+$rh)*$iiw+$rx]
                                  - $ii[$ry*$iiw+$rx+$rw])*$wt;
                        $rect_sum += $r_sum;
                    }
                    $rect_sum *= $inv_area;
                    $current_node = null;

                    if ($rect_sum >= $node_thresh*$vnorm) {
                        if ($rightidx == -1) {
                            $tree_sum = $rightval;
                        } else {
                            $current_node = $tree[$rightidx];
                        }
                    } else {
                        if ($leftidx == -1) {
                            $tree_sum = $leftval;
                        } else {
                            $current_node = $tree[$leftidx];
                        }
                    }
                }
                $stage_sum += $tree_sum;
            }
            if ($stage_sum < $stage_thresh) {
                return false;
            }
        }
        return true;
    }
}

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

<?php

class FaceDetector

{

protected $detection_data;

protected $canvas;

protected $face;

private $reduced_canvas;

public function __construct($detection_file = 'detection.dat')

{

if (is_file($detection_file)) {

$this->detection_data = unserialize(file_get_contents($detection_file));

} else {

throw new Exception("Couldn't load detection data");

}

public function faceDetect($file)

{

if (is_resource($file)) {

$this->canvas = $file;

} elseif (is_file($file)) {

$this->canvas = imagecreatefromjpeg($file);

} else {

throw new Exception("Can not load $file");

}

$im_width = imagesx($this->canvas);

$im_height = imagesy($this->canvas);

//Resample before detection?

$ratio = 0;

$diff_width = 320 - $im_width;

$diff_height = 240 - $im_height;

if ($diff_width > $diff_height) {

$ratio = $im_width / 320;

} else {

$ratio = $im_height / 240;

}

if ($ratio != 0) {

$this->reduced_canvas = imagecreatetruecolor($im_width / $ratio, $im_height / $ratio);

imagecopyresampled(

$this->reduced_canvas,

$this->canvas,

$im_width / $ratio,

$im_height / $ratio,

$im_width,

$im_height

);

$stats = $this->getImgStats($this->reduced_canvas);

$this->face = $this->doDetectGreedyBigToSmall(

$stats['ii'],

$stats['ii2'],

$stats['width'],

$stats['height']

);

if ($this->face['w'] > 0) {

$this->face['x'] *= $ratio;

$this->face['y'] *= $ratio;

$this->face['w'] *= $ratio;

}

} else {

$stats = $this->getImgStats($this->canvas);

$this->face = $this->doDetectGreedyBigToSmall(

$stats['ii'],

$stats['ii2'],

$stats['width'],

$stats['height']

);

}

return ($this->face['w'] > 0);

}

public function toJpeg()

{

$color = imagecolorallocate($this->canvas, 255, 0, 0); //red

imagerectangle(

$this->canvas,

$this->face['x'],

$this->face['y'],

$this->face['x']+$this->face['w'],

$this->face['y']+ $this->face['w'],

$color

);

header('Content-type: image/jpeg');

imagejpeg($this->canvas);

}

public function toJson()

{

return json_encode($this->face);

}

public function getFace()

{

return $this->face;

}

protected function getImgStats($canvas)

{

$image_width = imagesx($canvas);

$image_height = imagesy($canvas);

$iis = $this->computeII($canvas, $image_width, $image_height);

return array(

'width' => $image_width,

'height' => $image_height,

'ii' => $iis['ii'],

'ii2' => $iis['ii2']

);

}

protected function computeII($canvas, $image_width, $image_height)

{

$ii_w = $image_width+1;

$ii_h = $image_height+1;

$ii = array();

$ii2 = array();

for ($i=0; $i<$ii_w; $i++) {

$ii[$i] = 0;

$ii2[$i] = 0;

}

for ($i=1; $i<$ii_h-1; $i++) {

$ii[$i*$ii_w] = 0;

$ii2[$i*$ii_w] = 0;

$rowsum = 0;

$rowsum2 = 0;

for ($j=1; $j<$ii_w-1; $j++) {

$rgb = ImageColorAt($canvas, $j, $i);

$red = ($rgb >> 16) & 0xFF;

$green = ($rgb >> 8) & 0xFF;

$blue = $rgb & 0xFF;

$grey = (0.2989*$red + 0.587*$green + 0.114*$blue)>>0; // this is what matlab uses

$rowsum += $grey;

$rowsum2 += $grey*$grey;

$ii_above = ($i-1)*$ii_w + $j;

$ii_this = $i*$ii_w + $j;

$ii[$ii_this] = $ii[$ii_above] + $rowsum;

$ii2[$ii_this] = $ii2[$ii_above] + $rowsum2;

}

return array('ii'=>$ii, 'ii2' => $ii2);

}

protected function doDetectGreedyBigToSmall($ii, $ii2, $width, $height)

{

$s_w = $width/20.0;

$s_h = $height/20.0;

$start_scale = $s_h < $s_w ? $s_h : $s_w;

$scale_update = 1 / 1.2;

for ($scale = $start_scale; $scale > 1; $scale *= $scale_update) {

$w = (20*$scale) >> 0;

$endx = $width - $w - 1;

$endy = $height - $w - 1;

$step = max($scale, 2) >> 0;

$inv_area = 1 / ($w*$w);

for ($y = 0; $y < $endy; $y += $step) {

for ($x = 0; $x < $endx; $x += $step) {

$passed = $this->detectOnSubImage($x, $y, $scale, $ii, $ii2, $w, $width+1, $inv_area);

if ($passed) {

return array('x'=>$x, 'y'=>$y, 'w'=>$w);

}

} // end x

} // end y

} // end scale

return null;

}

protected function detectOnSubImage($x, $y, $scale, $ii, $ii2, $w, $iiw, $inv_area)

{

$mean = ($ii[($y+$w)*$iiw + $x + $w] + $ii[$y*$iiw+$x] - $ii[($y+$w)*$iiw+$x] - $ii[$y*$iiw+$x+$w])*$inv_area;

$vnorm = ($ii2[($y+$w)*$iiw + $x + $w]

+ $ii2[$y*$iiw+$x]

- $ii2[($y+$w)*$iiw+$x]

- $ii2[$y*$iiw+$x+$w])*$inv_area - ($mean*$mean);

$vnorm = $vnorm > 1 ? sqrt($vnorm) : 1;

$passed = true;

for ($i_stage = 0; $i_stage < count($this->detection_data); $i_stage++) {

$stage = $this->detection_data[$i_stage];

$trees = $stage[0];

$stage_thresh = $stage[1];

$stage_sum = 0;

for ($i_tree = 0; $i_tree < count($trees); $i_tree++) {

$tree = $trees[$i_tree];

$current_node = $tree[0];

$tree_sum = 0;

while ($current_node != null) {

$vals = $current_node[0];

$node_thresh = $vals[0];

$leftval = $vals[1];

$rightval = $vals[2];

$leftidx = $vals[3];

$rightidx = $vals[4];

$rects = $current_node[1];

$rect_sum = 0;

for ($i_rect = 0; $i_rect < count($rects); $i_rect++) {

$s = $scale;

$rect = $rects[$i_rect];

$rx = ($rect[0]*$s+$x)>>0;

$ry = ($rect[1]*$s+$y)>>0;

$rw = ($rect[2]*$s)>>0;

$rh = ($rect[3]*$s)>>0;

$wt = $rect[4];

$r_sum = ($ii[($ry+$rh)*$iiw + $rx + $rw]

+ $ii[$ry*$iiw+$rx]

- $ii[($ry+$rh)*$iiw+$rx]

- $ii[$ry*$iiw+$rx+$rw])*$wt;

$rect_sum += $r_sum;

}

$rect_sum *= $inv_area;

$current_node = null;

if ($rect_sum >= $node_thresh*$vnorm) {

if ($rightidx == -1) {

$tree_sum = $rightval;

} else {

$current_node = $tree[$rightidx];

}

} else {

if ($leftidx == -1) {

$tree_sum = $leftval;

} else {

$current_node = $tree[$leftidx];

}

$stage_sum += $tree_sum;

}

if ($stage_sum < $stage_thresh) {

return false;

}

return true;

}

And you simply use the class this way:

$detector = new Face_Detector('detection.dat'); $detector->face_detect('maurice_svay_150.jpg');    
$detector->toJpeg();

1 2	$detector = new Face_Detector('detection.dat'); $detector->face_detect('maurice_svay_150.jpg'); $detector->toJpeg();

Which gives the following result:

The code requires GD and is a bit slow but should work on most PHP servers. You’ll also need the data file: http://svay.com/experiences/face-detection/detection.dat. Let me know if you have ideas for improving the code 🙂

PHP

Face detection in PHP

UX & UI Design

Web Development

Mobile App Development

Ecommerce Development

Digital Marketing

Hire Dedicated Developer