CharacterEncodingDetector.cpp revision 1392eb3d1802e9f894f87d7a7387207d1b6faca1
1/* 2 * Copyright (C) 2013 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17//#define LOG_NDEBUG 0 18#define LOG_TAG "CharacterEncodingDector" 19#include <utils/Log.h> 20 21#include "CharacterEncodingDetector.h" 22#include "CharacterEncodingDetectorTables.h" 23 24#include "utils/Vector.h" 25#include "StringArray.h" 26 27#include "unicode/ucnv.h" 28#include "unicode/ucsdet.h" 29#include "unicode/ustring.h" 30 31namespace android { 32 33CharacterEncodingDetector::CharacterEncodingDetector() { 34 35 UErrorCode status = U_ZERO_ERROR; 36 mUtf8Conv = ucnv_open("UTF-8", &status); 37 if (U_FAILURE(status)) { 38 ALOGE("could not create UConverter for UTF-8"); 39 mUtf8Conv = NULL; 40 } 41} 42 43CharacterEncodingDetector::~CharacterEncodingDetector() { 44 ucnv_close(mUtf8Conv); 45} 46 47void CharacterEncodingDetector::addTag(const char *name, const char *value) { 48 mNames.push_back(name); 49 mValues.push_back(value); 50} 51 52size_t CharacterEncodingDetector::size() { 53 return mNames.size(); 54} 55 56status_t CharacterEncodingDetector::getTag(int index, const char **name, const char**value) { 57 if (index >= mNames.size()) { 58 return BAD_VALUE; 59 } 60 61 *name = mNames.getEntry(index); 62 *value = mValues.getEntry(index); 63 return OK; 64} 65 66static bool isPrintableAscii(const char *value, size_t len) { 67 for (size_t i = 0; i < len; i++) { 68 if ((value[i] & 0x80) || value[i] < 0x20 || value[i] == 0x7f) { 69 return false; 70 } 71 } 72 return true; 73} 74 75void CharacterEncodingDetector::detectAndConvert() { 76 77 int size = mNames.size(); 78 ALOGV("%d tags before conversion", size); 79 for (int i = 0; i < size; i++) { 80 ALOGV("%s: %s", mNames.getEntry(i), mValues.getEntry(i)); 81 } 82 83 if (size && mUtf8Conv) { 84 85 UErrorCode status = U_ZERO_ERROR; 86 UCharsetDetector *csd = ucsdet_open(&status); 87 const UCharsetMatch *ucm; 88 89 // try combined detection of artist/album/title etc. 90 char buf[1024]; 91 buf[0] = 0; 92 int idx; 93 bool allprintable = true; 94 for (int i = 0; i < size; i++) { 95 const char *name = mNames.getEntry(i); 96 const char *value = mValues.getEntry(i); 97 if (!isPrintableAscii(value, strlen(value)) && ( 98 !strcmp(name, "artist") || 99 !strcmp(name, "albumartist") || 100 !strcmp(name, "composer") || 101 !strcmp(name, "genre") || 102 !strcmp(name, "album") || 103 !strcmp(name, "title"))) { 104 strlcat(buf, value, sizeof(buf)); 105 // separate tags by space so ICU's ngram detector can do its job 106 strlcat(buf, " ", sizeof(buf)); 107 allprintable = false; 108 } 109 } 110 111 const char *combinedenc = "UTF-8"; 112 if (allprintable) { 113 // since 'buf' is empty, ICU would return a UTF-8 matcher with low confidence, so 114 // no need to even call it 115 ALOGV("all tags are printable, assuming ascii (%d)", strlen(buf)); 116 } else { 117 ucsdet_setText(csd, buf, strlen(buf), &status); 118 int32_t matches; 119 const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status); 120 bool goodmatch = true; 121 const UCharsetMatch* bestCombinedMatch = getPreferred(buf, strlen(buf), 122 ucma, matches, &goodmatch); 123 124 if (!goodmatch && strlen(buf) < 20) { 125 ALOGV("not a good match, trying with more data"); 126 // This string might be too short for ICU to do anything useful with. 127 // (real world example: "Björk" in ISO-8859-1 might be detected as GB18030, because 128 // the ISO detector reports a confidence of 0, while the GB18030 detector reports 129 // a confidence of 10 with no invalid characters) 130 // Append artist, album and title if they were previously omitted because they 131 // were printable ascii. 132 bool added = false; 133 for (int i = 0; i < size; i++) { 134 const char *name = mNames.getEntry(i); 135 const char *value = mValues.getEntry(i); 136 if (isPrintableAscii(value, strlen(value)) && ( 137 !strcmp(name, "artist") || 138 !strcmp(name, "album") || 139 !strcmp(name, "title"))) { 140 strlcat(buf, value, sizeof(buf)); 141 strlcat(buf, " ", sizeof(buf)); 142 added = true; 143 } 144 } 145 if (added) { 146 ucsdet_setText(csd, buf, strlen(buf), &status); 147 ucma = ucsdet_detectAll(csd, &matches, &status); 148 bestCombinedMatch = getPreferred(buf, strlen(buf), 149 ucma, matches, &goodmatch); 150 if (!goodmatch) { 151 ALOGV("still not a good match after adding printable tags"); 152 } 153 } else { 154 ALOGV("no printable tags to add"); 155 } 156 } 157 158 if (bestCombinedMatch != NULL) { 159 combinedenc = ucsdet_getName(bestCombinedMatch, &status); 160 } 161 } 162 163 for (int i = 0; i < size; i++) { 164 const char *name = mNames.getEntry(i); 165 uint8_t* src = (uint8_t *)mValues.getEntry(i); 166 int len = strlen((char *)src); 167 uint8_t* dest = src; 168 169 ALOGV("@@@ checking %s", name); 170 const char *s = mValues.getEntry(i); 171 int32_t inputLength = strlen(s); 172 const char *enc; 173 174 if (!allprintable && (!strcmp(name, "artist") || 175 !strcmp(name, "albumartist") || 176 !strcmp(name, "composer") || 177 !strcmp(name, "genre") || 178 !strcmp(name, "album") || 179 !strcmp(name, "title"))) { 180 // use encoding determined from the combination of artist/album/title etc. 181 enc = combinedenc; 182 } else { 183 if (isPrintableAscii(s, inputLength)) { 184 enc = "UTF-8"; 185 ALOGV("@@@@ %s is ascii", mNames.getEntry(i)); 186 } else { 187 ucsdet_setText(csd, s, inputLength, &status); 188 ucm = ucsdet_detect(csd, &status); 189 if (!ucm) { 190 mValues.setEntry(i, "???"); 191 continue; 192 } 193 enc = ucsdet_getName(ucm, &status); 194 ALOGV("@@@@ recognized charset: %s for %s confidence %d", 195 enc, mNames.getEntry(i), ucsdet_getConfidence(ucm, &status)); 196 } 197 } 198 199 if (strcmp(enc,"UTF-8") != 0) { 200 // only convert if the source encoding isn't already UTF-8 201 ALOGV("@@@ using converter %s for %s", enc, mNames.getEntry(i)); 202 UConverter *conv = ucnv_open(enc, &status); 203 if (U_FAILURE(status)) { 204 ALOGE("could not create UConverter for %s", enc); 205 continue; 206 } 207 208 // convert from native encoding to UTF-8 209 const char* source = mValues.getEntry(i); 210 int targetLength = len * 3 + 1; 211 char* buffer = new char[targetLength]; 212 // don't normally check for NULL, but in this case targetLength may be large 213 if (!buffer) 214 break; 215 char* target = buffer; 216 217 ucnv_convertEx(mUtf8Conv, conv, &target, target + targetLength, 218 &source, source + strlen(source), 219 NULL, NULL, NULL, NULL, TRUE, TRUE, &status); 220 221 if (U_FAILURE(status)) { 222 ALOGE("ucnv_convertEx failed: %d", status); 223 mValues.setEntry(i, "???"); 224 } else { 225 // zero terminate 226 *target = 0; 227 mValues.setEntry(i, buffer); 228 } 229 230 delete[] buffer; 231 232 ucnv_close(conv); 233 } 234 } 235 236 for (int i = size - 1; i >= 0; --i) { 237 if (strlen(mValues.getEntry(i)) == 0) { 238 ALOGV("erasing %s because entry is empty", mNames.getEntry(i)); 239 mNames.erase(i); 240 mValues.erase(i); 241 } 242 } 243 244 ucsdet_close(csd); 245 } 246} 247 248/* 249 * When ICU detects multiple encoding matches, apply additional heuristics to determine 250 * which one is the best match, since ICU can't always be trusted to make the right choice. 251 * 252 * What this method does is: 253 * - decode the input using each of the matches found 254 * - recalculate the starting confidence level for multibyte encodings using a different 255 * algorithm and larger frequent character lists than ICU 256 * - devalue encoding where the conversion contains unlikely characters (symbols, reserved, etc) 257 * - pick the highest match 258 * - signal to the caller whether this match is considered good: confidence > 15, and confidence 259 * delta with the next runner up > 15 260 */ 261const UCharsetMatch *CharacterEncodingDetector::getPreferred( 262 const char *input, size_t len, 263 const UCharsetMatch** ucma, size_t nummatches, 264 bool *goodmatch) { 265 266 *goodmatch = false; 267 Vector<const UCharsetMatch*> matches; 268 UErrorCode status = U_ZERO_ERROR; 269 270 ALOGV("%d matches", nummatches); 271 for (size_t i = 0; i < nummatches; i++) { 272 const char *encname = ucsdet_getName(ucma[i], &status); 273 int confidence = ucsdet_getConfidence(ucma[i], &status); 274 ALOGV("%d: %s %d", i, encname, confidence); 275 matches.push_back(ucma[i]); 276 } 277 278 size_t num = matches.size(); 279 if (num == 0) { 280 return NULL; 281 } 282 if (num == 1) { 283 int confidence = ucsdet_getConfidence(matches[0], &status); 284 if (confidence > 15) { 285 *goodmatch = true; 286 } 287 return matches[0]; 288 } 289 290 ALOGV("considering %d matches", num); 291 292 // keep track of how many "special" characters result when converting the input using each 293 // encoding 294 Vector<int> newconfidence; 295 for (size_t i = 0; i < num; i++) { 296 const uint16_t *freqdata = NULL; 297 float freqcoverage = 0; 298 status = U_ZERO_ERROR; 299 const char *encname = ucsdet_getName(matches[i], &status); 300 int confidence = ucsdet_getConfidence(matches[i], &status); 301 if (!strcmp("GB18030", encname)) { 302 freqdata = frequent_zhCN; 303 freqcoverage = frequent_zhCN_coverage; 304 } else if (!strcmp("Big5", encname)) { 305 freqdata = frequent_zhTW; 306 freqcoverage = frequent_zhTW_coverage; 307 } else if (!strcmp("EUC-KR", encname)) { 308 freqdata = frequent_ko; 309 freqcoverage = frequent_ko_coverage; 310 } else if (!strcmp("EUC-JP", encname)) { 311 freqdata = frequent_ja; 312 freqcoverage = frequent_ja_coverage; 313 } else if (!strcmp("Shift_JIS", encname)) { 314 freqdata = frequent_ja; 315 freqcoverage = frequent_ja_coverage; 316 } 317 318 ALOGV("%d: %s %d", i, encname, confidence); 319 UConverter *conv = ucnv_open(encname, &status); 320 const char *source = input; 321 const char *sourceLimit = input + len; 322 status = U_ZERO_ERROR; 323 int demerit = 0; 324 int frequentchars = 0; 325 int totalchars = 0; 326 while (true) { 327 // demerit the current encoding for each "special" character found after conversion. 328 // The amount of demerit is somewhat arbitrarily chosen. 329 int inchar; 330 if (source != sourceLimit) { 331 inchar = (source[0] << 8) + source[1]; 332 } 333 UChar32 c = ucnv_getNextUChar(conv, &source, sourceLimit, &status); 334 if (!U_SUCCESS(status)) { 335 break; 336 } 337 if (c < 0x20 || (c >= 0x7f && c <= 0x009f)) { 338 ALOGV("control character %x", c); 339 demerit += 100; 340 } else if ((c >= 0xa0 && c <= 0xbe) // symbols, superscripts 341 || (c == 0xd7) || (c == 0xf7) // multiplication and division signs 342 || (c >= 0x2000 && c <= 0x209f)) { // punctuation, superscripts 343 ALOGV("unlikely character %x", c); 344 demerit += 10; 345 } else if (c >= 0xe000 && c <= 0xf8ff) { 346 ALOGV("private use character %x", c); 347 demerit += 30; 348 } else if (c >= 0x2190 && c <= 0x2bff) { 349 // this range comprises various symbol ranges that are unlikely to appear in 350 // music file metadata. 351 ALOGV("symbol %x", c); 352 demerit += 10; 353 } else if (c == 0xfffd) { 354 ALOGV("replacement character"); 355 demerit += 50; 356 } else if (c >= 0xfff0 && c <= 0xfffc) { 357 ALOGV("unicode special %x", c); 358 demerit += 50; 359 } else if (freqdata != NULL) { 360 totalchars++; 361 if (isFrequent(freqdata, c)) { 362 frequentchars++; 363 } 364 } 365 } 366 if (freqdata != NULL && totalchars != 0) { 367 int myconfidence = 10 + float((100 * frequentchars) / totalchars) / freqcoverage; 368 ALOGV("ICU confidence: %d, my confidence: %d (%d %d)", confidence, myconfidence, 369 totalchars, frequentchars); 370 if (myconfidence > 100) myconfidence = 100; 371 if (myconfidence < 0) myconfidence = 0; 372 confidence = myconfidence; 373 } 374 ALOGV("%d-%d=%d", confidence, demerit, confidence - demerit); 375 newconfidence.push_back(confidence - demerit); 376 ucnv_close(conv); 377 if (i == 0 && (confidence - demerit) == 100) { 378 // no need to check any further, we'll end up using this match anyway 379 break; 380 } 381 } 382 383 // find match with highest confidence after adjusting for unlikely characters 384 int highest = newconfidence[0]; 385 size_t highestidx = 0; 386 int runnerup = -10000; 387 int runnerupidx = -10000; 388 num = newconfidence.size(); 389 for (size_t i = 1; i < num; i++) { 390 if (newconfidence[i] > highest) { 391 runnerup = highest; 392 runnerupidx = highestidx; 393 highest = newconfidence[i]; 394 highestidx = i; 395 } else if (newconfidence[i] > runnerup){ 396 runnerup = newconfidence[i]; 397 runnerupidx = i; 398 } 399 } 400 status = U_ZERO_ERROR; 401 ALOGV("selecting: '%s' w/ %d confidence", 402 ucsdet_getName(matches[highestidx], &status), highest); 403 if (runnerupidx < 0) { 404 ALOGV("no runner up"); 405 if (highest > 15) { 406 *goodmatch = true; 407 } 408 } else { 409 ALOGV("runner up: '%s' w/ %d confidence", 410 ucsdet_getName(matches[runnerupidx], &status), runnerup); 411 if ((highest - runnerup) > 15) { 412 *goodmatch = true; 413 } 414 } 415 return matches[highestidx]; 416} 417 418 419bool CharacterEncodingDetector::isFrequent(const uint16_t *values, uint32_t c) { 420 421 int start = 0; 422 int end = 511; // All the tables have 512 entries 423 int mid = (start+end)/2; 424 425 while(start <= end) { 426 if(c == values[mid]) { 427 return true; 428 } else if (c > values[mid]) { 429 start = mid + 1; 430 } else { 431 end = mid - 1; 432 } 433 434 mid = (start + end) / 2; 435 } 436 437 return false; 438} 439 440 441} // namespace android 442