Code Coverage |
||||||||||
Lines |
Functions and Methods |
Classes and Traits |
||||||||
| Total | |
97.00% |
97 / 100 |
|
87.50% |
7 / 8 |
CRAP | |
0.00% |
0 / 1 |
| BasicOcr | |
97.00% |
97 / 100 |
|
87.50% |
7 / 8 |
46 | |
0.00% |
0 / 1 |
| trainWith | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
1 | |||
| readImages | |
100.00% |
20 / 20 |
|
100.00% |
1 / 1 |
15 | |||
| readLabels | |
100.00% |
15 / 15 |
|
100.00% |
1 / 1 |
11 | |||
| kNearest | |
100.00% |
15 / 15 |
|
100.00% |
1 / 1 |
4 | |||
| getDistances | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
2 | |||
| imagesToMNIST | |
90.91% |
30 / 33 |
|
0.00% |
0 / 1 |
9.06 | |||
| labelsToMNIST | |
100.00% |
7 / 7 |
|
100.00% |
1 / 1 |
3 | |||
| matchImage | |
100.00% |
2 / 2 |
|
100.00% |
1 / 1 |
1 | |||
| 1 | <?php |
| 2 | /** |
| 3 | * Jingga |
| 4 | * |
| 5 | * PHP Version 8.1 |
| 6 | * |
| 7 | * @package phpOMS\Ai\Ocr |
| 8 | * @copyright Dennis Eichhorn |
| 9 | * @license OMS License 2.0 |
| 10 | * @version 1.0.0 |
| 11 | * @link https://jingga.app |
| 12 | */ |
| 13 | declare(strict_types=1); |
| 14 | |
| 15 | namespace phpOMS\Ai\Ocr; |
| 16 | |
| 17 | use phpOMS\Math\Topology\MetricsND; |
| 18 | use phpOMS\System\File\PathException; |
| 19 | |
| 20 | /** |
| 21 | * Basic OCR implementation for MNIST data |
| 22 | * |
| 23 | * @package phpOMS\Ai\Ocr |
| 24 | * @license OMS License 2.0 |
| 25 | * @link https://jingga.app |
| 26 | * @since 1.0.0 |
| 27 | */ |
| 28 | final class BasicOcr |
| 29 | { |
| 30 | /** |
| 31 | * Dataset on which the OCR is trained on. |
| 32 | * |
| 33 | * The data needs to be MNIST data. |
| 34 | * |
| 35 | * @var array |
| 36 | * @since 1.0.0 |
| 37 | */ |
| 38 | private array $Xtrain = []; |
| 39 | |
| 40 | /** |
| 41 | * Resultset on which the OCR is trained on. |
| 42 | * |
| 43 | * These are the actual values for the Xtrain data and must therefore have the same dimension. |
| 44 | * |
| 45 | * The labels need to be MNIST labels. |
| 46 | * |
| 47 | * @var array |
| 48 | * @since 1.0.0 |
| 49 | */ |
| 50 | private array $ytrain = []; |
| 51 | |
| 52 | /** |
| 53 | * Train OCR with data and result/labels |
| 54 | * |
| 55 | * @param string $dataPath Impage path to read |
| 56 | * @param string $labelPath Label path to read |
| 57 | * @param int $limit Limit (0 = unlimited) |
| 58 | * |
| 59 | * @return void |
| 60 | * |
| 61 | * @since 1.0.0 |
| 62 | */ |
| 63 | public function trainWith(string $dataPath, string $labelPath, int $limit = 0) : void |
| 64 | { |
| 65 | $Xtrain = $this->readImages($dataPath, $limit); |
| 66 | $ytrain = $this->readLabels($labelPath, $limit); |
| 67 | |
| 68 | $this->Xtrain = \array_merge($this->Xtrain, $Xtrain); |
| 69 | $this->ytrain = \array_merge($this->ytrain, $ytrain); |
| 70 | } |
| 71 | |
| 72 | /** |
| 73 | * Read image from path |
| 74 | * |
| 75 | * @param string $path Image to read |
| 76 | * @param int $limit Limit |
| 77 | * |
| 78 | * @return array |
| 79 | * |
| 80 | * @throws PathException |
| 81 | * |
| 82 | * @since 1.0.0 |
| 83 | */ |
| 84 | private function readImages(string $path, int $limit = 0) : array |
| 85 | { |
| 86 | if (!\is_file($path)) { |
| 87 | throw new PathException($path); |
| 88 | } |
| 89 | |
| 90 | $fp = \fopen($path, 'r'); |
| 91 | if ($fp === false) { |
| 92 | throw new PathException($path); // @codeCoverageIgnore |
| 93 | } |
| 94 | |
| 95 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 96 | return []; // @codeCoverageIgnore |
| 97 | } |
| 98 | |
| 99 | // $magicNumber = $unpack[1]; |
| 100 | // 2051 === image data (should always be this) |
| 101 | // 2049 === label data |
| 102 | |
| 103 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 104 | return []; // @codeCoverageIgnore |
| 105 | } |
| 106 | $numberOfImages = $unpack[1]; |
| 107 | |
| 108 | if ($limit > 0) { |
| 109 | $numberOfImages = \min($numberOfImages, $limit); |
| 110 | } |
| 111 | |
| 112 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 113 | return []; // @codeCoverageIgnore |
| 114 | } |
| 115 | |
| 116 | /** @var int<0, max> $numberOfRows */ |
| 117 | $numberOfRows = (int) $unpack[1]; |
| 118 | |
| 119 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 120 | return []; // @codeCoverageIgnore |
| 121 | } |
| 122 | |
| 123 | /** @var int<0, max> $numberOfColumns */ |
| 124 | $numberOfColumns = (int) $unpack[1]; |
| 125 | |
| 126 | $images = []; |
| 127 | for ($i = 0; $i < $numberOfImages; ++$i) { |
| 128 | if (($read = \fread($fp, $numberOfRows * $numberOfColumns)) === false |
| 129 | || ($unpack = \unpack('C*', $read)) === false |
| 130 | ) { |
| 131 | return []; // @codeCoverageIgnore |
| 132 | } |
| 133 | |
| 134 | $images[] = \array_values($unpack); |
| 135 | } |
| 136 | |
| 137 | \fclose($fp); |
| 138 | |
| 139 | return $images; |
| 140 | } |
| 141 | |
| 142 | /** |
| 143 | * Read labels from from path |
| 144 | * |
| 145 | * @param string $path Labels path |
| 146 | * @param int $limit Limit |
| 147 | * |
| 148 | * @return array |
| 149 | * |
| 150 | * @throws PathException |
| 151 | * |
| 152 | * @since 1.0.0 |
| 153 | */ |
| 154 | private function readLabels(string $path, int $limit = 0) : array |
| 155 | { |
| 156 | if (!\is_file($path)) { |
| 157 | throw new PathException($path); |
| 158 | } |
| 159 | |
| 160 | $fp = \fopen($path, 'r'); |
| 161 | if ($fp === false) { |
| 162 | throw new PathException($path); // @codeCoverageIgnore |
| 163 | } |
| 164 | |
| 165 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 166 | return []; // @codeCoverageIgnore |
| 167 | } |
| 168 | |
| 169 | // $magicNumber = $unpack[1]; |
| 170 | // 2051 === image data |
| 171 | // 2049 === label data (should always be this) |
| 172 | |
| 173 | if (($read = \fread($fp, 4)) === false || ($unpack = \unpack('N', $read)) === false) { |
| 174 | return []; // @codeCoverageIgnore |
| 175 | } |
| 176 | $numberOfLabels = $unpack[1]; |
| 177 | |
| 178 | if ($limit > 0) { |
| 179 | $numberOfLabels = \min($numberOfLabels, $limit); |
| 180 | } |
| 181 | |
| 182 | $labels = []; |
| 183 | for ($i = 0; $i < $numberOfLabels; ++$i) { |
| 184 | if (($read = \fread($fp, 1)) === false || ($unpack = \unpack('C', $read)) === false) { |
| 185 | return []; // @codeCoverageIgnore |
| 186 | } |
| 187 | $labels[] = $unpack[1]; |
| 188 | } |
| 189 | |
| 190 | \fclose($fp); |
| 191 | |
| 192 | return $labels; |
| 193 | } |
| 194 | |
| 195 | /** |
| 196 | * Find the k-nearest matches for test data |
| 197 | * |
| 198 | * @param array $Xtrain Image data used for training |
| 199 | * @param array $ytrain Labels associated with the trained data |
| 200 | * @param array $Xtest Image data from the image to categorize |
| 201 | * @param int $k Amount of best fits that should be found |
| 202 | */ |
| 203 | private function kNearest(array $Xtrain, array $ytrain, array $Xtest, int $k = 3) : array |
| 204 | { |
| 205 | $predictedLabels = []; |
| 206 | foreach ($Xtest as $sample) { |
| 207 | $distances = $this->getDistances($Xtrain, $sample); |
| 208 | \asort($distances); |
| 209 | |
| 210 | $keys = \array_keys($distances); |
| 211 | |
| 212 | $candidateLabels = []; |
| 213 | for ($i = 0; $i < $k; ++$i) { |
| 214 | $candidateLabels[] = $ytrain[$keys[$i]]; |
| 215 | } |
| 216 | |
| 217 | // find best match |
| 218 | $countedCandidates = \array_count_values($candidateLabels); |
| 219 | |
| 220 | foreach ($candidateLabels as $i => $label) { |
| 221 | $predictedLabels[] = [ |
| 222 | 'label' => $label, |
| 223 | 'prob' => $countedCandidates[$label] / $k, |
| 224 | ]; |
| 225 | } |
| 226 | } |
| 227 | |
| 228 | return $predictedLabels; |
| 229 | } |
| 230 | |
| 231 | /** |
| 232 | * Fitting method in order to see how similar two datasets are. |
| 233 | * |
| 234 | * @param array $Xtrain Image data used for training |
| 235 | * @param array $sample Image data to compare against |
| 236 | * |
| 237 | * @return array |
| 238 | * |
| 239 | * @since 1.0.0 |
| 240 | */ |
| 241 | private function getDistances(array $Xtrain, array $sample) : array |
| 242 | { |
| 243 | $dist = []; |
| 244 | foreach ($Xtrain as $train) { |
| 245 | $dist[] = MetricsND::euclidean($train, $sample); |
| 246 | } |
| 247 | |
| 248 | return $dist; |
| 249 | } |
| 250 | |
| 251 | /** |
| 252 | * Create MNIST file from images |
| 253 | * |
| 254 | * @param string[] $images Images |
| 255 | * @param string $out Output file |
| 256 | * @param int $resolution Resolution of the iomages |
| 257 | * |
| 258 | * @return void |
| 259 | * |
| 260 | * @since 1.0.0 |
| 261 | */ |
| 262 | public static function imagesToMNIST(array $images, string $out, int $resolution) : void |
| 263 | { |
| 264 | $out = \fopen($out, 'wb'); |
| 265 | if ($out === false) { |
| 266 | return; // @codeCoverageIgnore |
| 267 | } |
| 268 | |
| 269 | \fwrite($out, \pack('N', 2051)); |
| 270 | \fwrite($out, \pack('N', \count($images))); |
| 271 | \fwrite($out, \pack('N', $resolution)); |
| 272 | \fwrite($out, \pack('N', $resolution)); |
| 273 | |
| 274 | $size = $resolution * $resolution; |
| 275 | |
| 276 | foreach ($images as $in) { |
| 277 | $inString = \file_get_contents($in); |
| 278 | if ($inString === false) { |
| 279 | continue; |
| 280 | } |
| 281 | |
| 282 | $im = \imagecreatefromstring($inString); |
| 283 | if ($im === false) { |
| 284 | continue; |
| 285 | } |
| 286 | |
| 287 | $new = \imagescale($im, $resolution, $resolution); |
| 288 | if ($new === false) { |
| 289 | continue; |
| 290 | } |
| 291 | |
| 292 | // Convert the image to grayscale and normalize the pixel values |
| 293 | $mnist = []; |
| 294 | for ($i = 0; $i < $resolution; ++$i) { |
| 295 | for ($j = 0; $j < $resolution; ++$j) { |
| 296 | $pixel = \imagecolorat($new, $j, $i); |
| 297 | $gray = \round( |
| 298 | ( |
| 299 | 0.299 * (($pixel >> 16) & 0xFF) |
| 300 | + 0.587 * (($pixel >> 8) & 0xFF) |
| 301 | + 0.114 * ($pixel & 0xFF) |
| 302 | ) / 255, |
| 303 | 3 |
| 304 | ); |
| 305 | |
| 306 | $mnist[] = $gray; |
| 307 | } |
| 308 | } |
| 309 | |
| 310 | for ($i = 0; $i < $size; ++$i) { |
| 311 | \fwrite($out, \pack('C', (int) \round($mnist[$i] * 255))); |
| 312 | } |
| 313 | } |
| 314 | |
| 315 | \fclose($out); |
| 316 | } |
| 317 | |
| 318 | /** |
| 319 | * Convert labels to MNIST format |
| 320 | * |
| 321 | * @param string[] $data Labels (one char per label) |
| 322 | * @param string $out Output path |
| 323 | * |
| 324 | * @return void |
| 325 | * |
| 326 | * @since 1.0.0 |
| 327 | */ |
| 328 | public static function labelsToMNIST(array $data, string $out) : void |
| 329 | { |
| 330 | // Only allows single char labels |
| 331 | $out = \fopen($out, 'wb'); |
| 332 | if ($out === false) { |
| 333 | return; // @codeCoverageIgnore |
| 334 | } |
| 335 | |
| 336 | \fwrite($out, \pack('N', 2049)); |
| 337 | \fwrite($out, \pack('N', \count($data))); |
| 338 | |
| 339 | foreach ($data as $e) { |
| 340 | \fwrite($out, \pack('C', $e)); |
| 341 | } |
| 342 | |
| 343 | \fclose($out); |
| 344 | } |
| 345 | |
| 346 | /** |
| 347 | * Categorize an unknown image |
| 348 | * |
| 349 | * @param string $path Path to the image to categorize/evaluate/match against the training data |
| 350 | * @param int $comparison Amount of comparisons |
| 351 | * @param int $limit Limit (0 = unlimited) |
| 352 | * |
| 353 | * @return array |
| 354 | * |
| 355 | * @since 1.0.0 |
| 356 | */ |
| 357 | public function matchImage(string $path, int $comparison = 3, int $limit = 0) : array |
| 358 | { |
| 359 | $Xtest = $this->readImages($path, $limit); |
| 360 | |
| 361 | return $this->kNearest($this->Xtrain, $this->ytrain, $Xtest, $comparison); |
| 362 | } |
| 363 | } |