1 #include <cmath> 2 3 #include "SkBitmap.h" 4 #include "skpdiff_util.h" 5 #include "SkPMetric.h" 6 #include "SkPMetricUtil_generated.h" 7 8 struct RGB { 9 float r, g, b; 10 }; 11 12 struct LAB { 13 float l, a, b; 14 }; 15 16 template<class T> 17 struct Image2D { 18 int width; 19 int height; 20 T* image; 21 22 Image2D(int w, int h) 23 : width(w), 24 height(h) { 25 SkASSERT(w > 0); 26 SkASSERT(h > 0); 27 image = SkNEW_ARRAY(T, w * h); 28 } 29 30 ~Image2D() { 31 SkDELETE_ARRAY(image); 32 } 33 34 void readPixel(int x, int y, T* pixel) const { 35 SkASSERT(x >= 0); 36 SkASSERT(y >= 0); 37 SkASSERT(x < width); 38 SkASSERT(y < height); 39 *pixel = image[y * width + x]; 40 } 41 42 T* getRow(int y) const { 43 return &image[y * width]; 44 } 45 46 void writePixel(int x, int y, const T& pixel) { 47 SkASSERT(x >= 0); 48 SkASSERT(y >= 0); 49 SkASSERT(x < width); 50 SkASSERT(y < height); 51 image[y * width + x] = pixel; 52 } 53 }; 54 55 typedef Image2D<float> ImageL; 56 typedef Image2D<RGB> ImageRGB; 57 typedef Image2D<LAB> ImageLAB; 58 59 template<class T> 60 struct ImageArray 61 { 62 int slices; 63 Image2D<T>** image; 64 65 ImageArray(int w, int h, int s) 66 : slices(s) { 67 SkASSERT(s > 0); 68 image = SkNEW_ARRAY(Image2D<T>*, s); 69 for (int sliceIndex = 0; sliceIndex < slices; sliceIndex++) { 70 image[sliceIndex] = SkNEW_ARGS(Image2D<T>, (w, h)); 71 } 72 } 73 74 ~ImageArray() { 75 for (int sliceIndex = 0; sliceIndex < slices; sliceIndex++) { 76 SkDELETE(image[sliceIndex]); 77 } 78 SkDELETE_ARRAY(image); 79 } 80 81 Image2D<T>* getLayer(int z) const { 82 SkASSERT(z >= 0); 83 SkASSERT(z < slices); 84 return image[z]; 85 } 86 }; 87 88 typedef ImageArray<float> ImageL3D; 89 90 91 #define MAT_ROW_MULT(rc,gc,bc) r*rc + g*gc + b*bc 92 93 static void adobergb_to_cielab(float r, float g, float b, LAB* lab) { 94 // Conversion of Adobe RGB to XYZ taken from from "Adobe RGB (1998) ColorImage Encoding" 95 // URL:http://www.adobe.com/digitalimag/pdfs/AdobeRGB1998.pdf 96 // Section: 4.3.5.3 97 // See Also: http://en.wikipedia.org/wiki/Adobe_rgb 98 float x = MAT_ROW_MULT(0.57667f, 0.18556f, 0.18823f); 99 float y = MAT_ROW_MULT(0.29734f, 0.62736f, 0.07529f); 100 float z = MAT_ROW_MULT(0.02703f, 0.07069f, 0.99134f); 101 102 // The following is the white point in XYZ, so it's simply the row wise addition of the above 103 // matrix. 104 const float xw = 0.5767f + 0.185556f + 0.188212f; 105 const float yw = 0.297361f + 0.627355f + 0.0752847f; 106 const float zw = 0.0270328f + 0.0706879f + 0.991248f; 107 108 // This is the XYZ color point relative to the white point 109 float f[3] = { x / xw, y / yw, z / zw }; 110 111 // Conversion from XYZ to LAB taken from 112 // http://en.wikipedia.org/wiki/CIELAB#Forward_transformation 113 for (int i = 0; i < 3; i++) { 114 if (f[i] >= 0.008856f) { 115 f[i] = SkPMetricUtil::get_cube_root(f[i]); 116 } else { 117 f[i] = 7.787f * f[i] + 4.0f / 29.0f; 118 } 119 } 120 lab->l = 116.0f * f[1] - 16.0f; 121 lab->a = 500.0f * (f[0] - f[1]); 122 lab->b = 200.0f * (f[1] - f[2]); 123 } 124 125 /// Converts a 8888 bitmap to LAB color space and puts it into the output 126 static bool bitmap_to_cielab(const SkBitmap* bitmap, ImageLAB* outImageLAB) { 127 SkBitmap bm8888; 128 if (bitmap->config() != SkBitmap::kARGB_8888_Config) { 129 if (!bitmap->copyTo(&bm8888, SkBitmap::kARGB_8888_Config)) { 130 return false; 131 } 132 bitmap = &bm8888; 133 } 134 135 int width = bitmap->width(); 136 int height = bitmap->height(); 137 SkASSERT(outImageLAB->width == width); 138 SkASSERT(outImageLAB->height == height); 139 140 bitmap->lockPixels(); 141 RGB rgb; 142 LAB lab; 143 for (int y = 0; y < height; y++) { 144 unsigned char* row = (unsigned char*)bitmap->getAddr(0, y); 145 for (int x = 0; x < width; x++) { 146 // Perform gamma correction which is assumed to be 2.2 147 rgb.r = SkPMetricUtil::get_gamma(row[x * 4 + 2]); 148 rgb.g = SkPMetricUtil::get_gamma(row[x * 4 + 1]); 149 rgb.b = SkPMetricUtil::get_gamma(row[x * 4 + 0]); 150 adobergb_to_cielab(rgb.r, rgb.g, rgb.b, &lab); 151 outImageLAB->writePixel(x, y, lab); 152 } 153 } 154 bitmap->unlockPixels(); 155 return true; 156 } 157 158 // From Barten SPIE 1989 159 static float contrast_sensitivity(float cyclesPerDegree, float luminance) { 160 float a = 440.0f * powf(1.0f + 0.7f / luminance, -0.2f); 161 float b = 0.3f * powf(1.0f + 100.0f / luminance, 0.15f); 162 return a * 163 cyclesPerDegree * 164 expf(-b * cyclesPerDegree) * 165 sqrtf(1.0f + 0.06f * expf(b * cyclesPerDegree)); 166 } 167 168 #if 0 169 // We're keeping these around for reference and in case the lookup tables are no longer desired. 170 // They are no longer called by any code in this file. 171 172 // From Daly 1993 173 static float visual_mask(float contrast) { 174 float x = powf(392.498f * contrast, 0.7f); 175 x = powf(0.0153f * x, 4.0f); 176 return powf(1.0f + x, 0.25f); 177 } 178 179 // From Ward Larson Siggraph 1997 180 static float threshold_vs_intensity(float adaptationLuminance) { 181 float logLum = log10f(adaptationLuminance); 182 float x; 183 if (logLum < -3.94f) { 184 x = -2.86f; 185 } else if (logLum < -1.44f) { 186 x = powf(0.405f * logLum + 1.6f, 2.18) - 2.86f; 187 } else if (logLum < -0.0184f) { 188 x = logLum - 0.395f; 189 } else if (logLum < 1.9f) { 190 x = powf(0.249f * logLum + 0.65f, 2.7f) - 0.72f; 191 } else { 192 x = logLum - 1.255f; 193 } 194 return powf(10.0f, x); 195 } 196 197 #endif 198 199 /// Simply takes the L channel from the input and puts it into the output 200 static void lab_to_l(const ImageLAB* imageLAB, ImageL* outImageL) { 201 for (int y = 0; y < imageLAB->height; y++) { 202 for (int x = 0; x < imageLAB->width; x++) { 203 LAB lab; 204 imageLAB->readPixel(x, y, &lab); 205 outImageL->writePixel(x, y, lab.l); 206 } 207 } 208 } 209 210 /// Convolves an image with the given filter in one direction and saves it to the output image 211 static void convolve(const ImageL* imageL, bool vertical, ImageL* outImageL) { 212 SkASSERT(imageL->width == outImageL->width); 213 SkASSERT(imageL->height == outImageL->height); 214 215 const float matrix[] = { 0.05f, 0.25f, 0.4f, 0.25f, 0.05f }; 216 const int matrixCount = sizeof(matrix) / sizeof(float); 217 const int radius = matrixCount / 2; 218 219 // Keep track of what rows are being operated on for quick access. 220 float* rowPtrs[matrixCount]; // Because matrixCount is constant, this won't create a VLA 221 for (int y = radius; y < matrixCount; y++) { 222 rowPtrs[y] = imageL->getRow(y - radius); 223 } 224 float* writeRow = outImageL->getRow(0); 225 226 for (int y = 0; y < imageL->height; y++) { 227 for (int x = 0; x < imageL->width; x++) { 228 float lSum = 0.0f; 229 for (int xx = -radius; xx <= radius; xx++) { 230 int nx = x; 231 int ny = y; 232 233 // We mirror at edges so that edge pixels that the filter weighting still makes 234 // sense. 235 if (vertical) { 236 ny += xx; 237 if (ny < 0) { 238 ny = -ny; 239 } 240 if (ny >= imageL->height) { 241 ny = imageL->height + (imageL->height - ny - 1); 242 } 243 } else { 244 nx += xx; 245 if (nx < 0) { 246 nx = -nx; 247 } 248 if (nx >= imageL->width) { 249 nx = imageL->width + (imageL->width - nx - 1); 250 } 251 } 252 253 float weight = matrix[xx + radius]; 254 lSum += rowPtrs[ny - y + radius][nx] * weight; 255 } 256 writeRow[x] = lSum; 257 } 258 // As we move down, scroll the row pointers down with us 259 for (int y = 0; y < matrixCount - 1; y++) 260 { 261 rowPtrs[y] = rowPtrs[y + 1]; 262 } 263 rowPtrs[matrixCount - 1] += imageL->width; 264 writeRow += imageL->width; 265 } 266 } 267 268 static double pmetric(const ImageLAB* baselineLAB, const ImageLAB* testLAB, int* poiCount) { 269 SkASSERT(baselineLAB); 270 SkASSERT(testLAB); 271 SkASSERT(poiCount); 272 273 int width = baselineLAB->width; 274 int height = baselineLAB->height; 275 int maxLevels = 0; 276 277 // Calculates how many levels to make by how many times the image can be divided in two 278 int smallerDimension = width < height ? width : height; 279 for ( ; smallerDimension > 1; smallerDimension /= 2) { 280 maxLevels++; 281 } 282 283 // We'll be creating new arrays with maxLevels - 2, and ImageL3D requires maxLevels > 0, 284 // so just return failure if we're less than 3. 285 if (maxLevels <= 2) { 286 return 0.0; 287 } 288 289 const float fov = SK_ScalarPI / 180.0f * 45.0f; 290 float contrastSensitivityMax = contrast_sensitivity(3.248f, 100.0f); 291 float pixelsPerDegree = width / (2.0f * tanf(fov * 0.5f) * 180.0f / SK_ScalarPI); 292 293 ImageL3D baselineL(width, height, maxLevels); 294 ImageL3D testL(width, height, maxLevels); 295 ImageL scratchImageL(width, height); 296 float* cyclesPerDegree = SkNEW_ARRAY(float, maxLevels); 297 float* thresholdFactorFrequency = SkNEW_ARRAY(float, maxLevels - 2); 298 float* contrast = SkNEW_ARRAY(float, maxLevels - 2); 299 300 lab_to_l(baselineLAB, baselineL.getLayer(0)); 301 lab_to_l(testLAB, testL.getLayer(0)); 302 303 // Compute cpd - Cycles per degree on the pyramid 304 cyclesPerDegree[0] = 0.5f * pixelsPerDegree; 305 for (int levelIndex = 1; levelIndex < maxLevels; levelIndex++) { 306 cyclesPerDegree[levelIndex] = cyclesPerDegree[levelIndex - 1] * 0.5f; 307 } 308 309 // Contrast sensitivity is based on image dimensions. Therefore it cannot be statically 310 // generated. 311 float* contrastSensitivityTable = SkNEW_ARRAY(float, maxLevels * 1000); 312 for (int levelIndex = 0; levelIndex < maxLevels; levelIndex++) { 313 for (int csLum = 0; csLum < 1000; csLum++) { 314 contrastSensitivityTable[levelIndex * 1000 + csLum] = 315 contrast_sensitivity(cyclesPerDegree[levelIndex], (float)csLum / 10.0f + 1e-5f); 316 } 317 } 318 319 // Compute G - The convolved lum for the baseline 320 for (int levelIndex = 1; levelIndex < maxLevels; levelIndex++) { 321 convolve(baselineL.getLayer(levelIndex - 1), false, &scratchImageL); 322 convolve(&scratchImageL, true, baselineL.getLayer(levelIndex)); 323 } 324 for (int levelIndex = 1; levelIndex < maxLevels; levelIndex++) { 325 convolve(testL.getLayer(levelIndex - 1), false, &scratchImageL); 326 convolve(&scratchImageL, true, testL.getLayer(levelIndex)); 327 } 328 329 // Compute F_freq - The elevation f 330 for (int levelIndex = 0; levelIndex < maxLevels - 2; levelIndex++) { 331 float cpd = cyclesPerDegree[levelIndex]; 332 thresholdFactorFrequency[levelIndex] = contrastSensitivityMax / 333 contrast_sensitivity(cpd, 100.0f); 334 } 335 336 // Calculate F 337 for (int y = 0; y < height; y++) { 338 for (int x = 0; x < width; x++) { 339 float lBaseline; 340 float lTest; 341 baselineL.getLayer(0)->readPixel(x, y, &lBaseline); 342 testL.getLayer(0)->readPixel(x, y, &lTest); 343 344 float avgLBaseline; 345 float avgLTest; 346 baselineL.getLayer(maxLevels - 1)->readPixel(x, y, &avgLBaseline); 347 testL.getLayer(maxLevels - 1)->readPixel(x, y, &avgLTest); 348 349 float lAdapt = 0.5f * (avgLBaseline + avgLTest); 350 if (lAdapt < 1e-5f) { 351 lAdapt = 1e-5f; 352 } 353 354 float contrastSum = 0.0f; 355 for (int levelIndex = 0; levelIndex < maxLevels - 2; levelIndex++) { 356 float baselineL0, baselineL1, baselineL2; 357 float testL0, testL1, testL2; 358 baselineL.getLayer(levelIndex + 0)->readPixel(x, y, &baselineL0); 359 testL. getLayer(levelIndex + 0)->readPixel(x, y, &testL0); 360 baselineL.getLayer(levelIndex + 1)->readPixel(x, y, &baselineL1); 361 testL. getLayer(levelIndex + 1)->readPixel(x, y, &testL1); 362 baselineL.getLayer(levelIndex + 2)->readPixel(x, y, &baselineL2); 363 testL. getLayer(levelIndex + 2)->readPixel(x, y, &testL2); 364 365 float baselineContrast1 = fabsf(baselineL0 - baselineL1); 366 float testContrast1 = fabsf(testL0 - testL1); 367 float numerator = (baselineContrast1 > testContrast1) ? 368 baselineContrast1 : testContrast1; 369 370 float baselineContrast2 = fabsf(baselineL2); 371 float testContrast2 = fabsf(testL2); 372 float denominator = (baselineContrast2 > testContrast2) ? 373 baselineContrast2 : testContrast2; 374 375 // Avoid divides by close to zero 376 if (denominator < 1e-5f) { 377 denominator = 1e-5f; 378 } 379 contrast[levelIndex] = numerator / denominator; 380 contrastSum += contrast[levelIndex]; 381 } 382 383 if (contrastSum < 1e-5f) { 384 contrastSum = 1e-5f; 385 } 386 387 float F = 0.0f; 388 for (int levelIndex = 0; levelIndex < maxLevels - 2; levelIndex++) { 389 float contrastSensitivity = contrastSensitivityTable[levelIndex * 1000 + 390 (int)(lAdapt * 10.0)]; 391 float mask = SkPMetricUtil::get_visual_mask(contrast[levelIndex] * 392 contrastSensitivity); 393 394 F += contrast[levelIndex] + 395 thresholdFactorFrequency[levelIndex] * mask / contrastSum; 396 } 397 398 if (F < 1.0f) { 399 F = 1.0f; 400 } 401 402 if (F > 10.0f) { 403 F = 10.0f; 404 } 405 406 407 bool isFailure = false; 408 if (fabsf(lBaseline - lTest) > F * SkPMetricUtil::get_threshold_vs_intensity(lAdapt)) { 409 isFailure = true; 410 } else { 411 LAB baselineColor; 412 LAB testColor; 413 baselineLAB->readPixel(x, y, &baselineColor); 414 testLAB->readPixel(x, y, &testColor); 415 float contrastA = baselineColor.a - testColor.a; 416 float contrastB = baselineColor.b - testColor.b; 417 float colorScale = 1.0f; 418 if (lAdapt < 10.0f) { 419 colorScale = lAdapt / 10.0f; 420 } 421 colorScale *= colorScale; 422 423 if ((contrastA * contrastA + contrastB * contrastB) * colorScale > F) 424 { 425 isFailure = true; 426 } 427 } 428 429 if (isFailure) { 430 (*poiCount)++; 431 } 432 } 433 } 434 435 SkDELETE_ARRAY(cyclesPerDegree); 436 SkDELETE_ARRAY(contrast); 437 SkDELETE_ARRAY(thresholdFactorFrequency); 438 SkDELETE_ARRAY(contrastSensitivityTable); 439 return 1.0 - (double)(*poiCount) / (width * height); 440 } 441 442 bool SkPMetric::diff(SkBitmap* baseline, SkBitmap* test, bool computeMask, Result* result) const { 443 double startTime = get_seconds(); 444 445 // Ensure the images are comparable 446 if (baseline->width() != test->width() || baseline->height() != test->height() || 447 baseline->width() <= 0 || baseline->height() <= 0) { 448 return false; 449 } 450 451 ImageLAB baselineLAB(baseline->width(), baseline->height()); 452 ImageLAB testLAB(baseline->width(), baseline->height()); 453 454 if (!bitmap_to_cielab(baseline, &baselineLAB) || !bitmap_to_cielab(test, &testLAB)) { 455 return true; 456 } 457 458 result->poiCount = 0; 459 result->result = pmetric(&baselineLAB, &testLAB, &result->poiCount); 460 result->timeElapsed = get_seconds() - startTime; 461 462 return true; 463 } 464