Spaces:
Sleeping
Sleeping
File size: 14,449 Bytes
900df0b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 | """
OmniFile AI Processor — Pattern Matching Engine
=================================================
Source: arabic-ocr-pro/ai/pattern_matcher.py
Uses SSIM (Structural Similarity Index) to compare new OCR word images
against stored patterns from the pattern database. When a match is found,
the stored correct label can replace the OCR output.
This enables the system to learn from user corrections: when a user
corrects "المملكك" to "المملكة", the corrected word image is stored
as a pattern. Future occurrences of similar-looking word images will
be automatically corrected.
"""
from __future__ import annotations
import logging
from typing import Optional
import cv2
import numpy as np
logger = logging.getLogger(__name__)
# Lazy import for skimage to avoid hard dependency at module load
_skimage_ssim = None
def _get_ssim_fn():
"""Lazily import and cache the SSIM function."""
global _skimage_ssim
if _skimage_ssim is None:
try:
from skimage.metrics import structural_similarity
_skimage_ssim = structural_similarity
except ImportError:
logger.warning(
"scikit-image not installed. SSIM matching unavailable. "
"Install with: pip install scikit-image"
)
_skimage_ssim = False # sentinel
return _skimage_ssim if _skimage_ssim else None
class PatternMatch:
"""Result of a pattern matching operation.
Attributes:
pattern_id: ID of the matched pattern in the database.
label: Correct text label from the matched pattern.
confidence: SSIM similarity score (0.0 to 1.0).
pattern_image: Decoded pattern image (numpy array).
"""
def __init__(
self,
pattern_id: int,
label: str,
confidence: float,
pattern_image: Optional[np.ndarray] = None,
) -> None:
"""Initialize a pattern match result.
Args:
pattern_id: Database ID of the matched pattern.
label: Correct text label.
confidence: Similarity confidence score.
pattern_image: Decoded image of the pattern.
"""
self.pattern_id = pattern_id
self.label = label
self.confidence = confidence
self.pattern_image = pattern_image
def __repr__(self) -> str:
return (
f"PatternMatch(id={self.pattern_id}, "
f"label='{self.label}', conf={self.confidence:.3f})"
)
class PatternMatcher:
"""SSIM-based pattern matching engine.
Compares word images against stored patterns to find matches
and suggest corrections based on previously learned corrections.
Attributes:
db: Pattern database instance.
threshold: Minimum SSIM score to consider a match valid.
_patterns_cache: In-memory cache of loaded patterns.
"""
def __init__(
self,
db: Optional["PatternDatabase"] = None, # type: ignore[type-arg]
db_path: str = "data/corrections.db",
threshold: float = 0.85,
) -> None:
"""Initialize the pattern matcher.
Args:
db: Pattern database instance. Creates a new one if None.
db_path: Path to the database (used only if db is None).
threshold: Minimum SSIM threshold for accepting matches.
"""
# Lazy import to avoid circular dependency
if db is None:
from modules.ai.pattern_db import PatternDatabase
db = PatternDatabase(db_path)
self.db = db
self.threshold = threshold
self._patterns_cache: list[dict] = []
# ------------------------------------------------------------------
# Loading patterns
# ------------------------------------------------------------------
def load_patterns(self, label_filter: Optional[str] = None) -> int:
"""Load patterns from the database into memory cache.
Loads pattern images and their labels for fast matching.
Should be called before batch matching operations.
Args:
label_filter: Optional label to filter patterns.
Returns:
Number of patterns loaded.
"""
patterns = self.db.get_patterns(label=label_filter, limit=5000)
self._patterns_cache = patterns
logger.info(f"Loaded {len(self._patterns_cache)} patterns for matching")
return len(self._patterns_cache)
# ------------------------------------------------------------------
# Matching
# ------------------------------------------------------------------
def match_word(
self,
word_image: np.ndarray,
label_filter: Optional[str] = None,
) -> Optional[PatternMatch]:
"""Find the best matching pattern for a word image.
Compares the input word image against all cached patterns
using SSIM and returns the best match above the threshold.
Args:
word_image: Cropped word image (grayscale or BGR).
label_filter: Optional label to filter search.
Returns:
PatternMatch if a match is found, None otherwise.
"""
ssim_fn = _get_ssim_fn()
if ssim_fn is None:
logger.debug("SSIM not available — skipping image matching")
return None
# Convert to grayscale if needed
if len(word_image.shape) == 3:
gray_word = cv2.cvtColor(word_image, cv2.COLOR_BGR2GRAY)
else:
gray_word = word_image.copy()
# Ensure patterns are loaded
if not self._patterns_cache:
self.load_patterns(label_filter)
best_match: Optional[PatternMatch] = None
best_score = -1.0
for pattern in self._patterns_cache:
# Filter by label if specified
if label_filter and pattern["label"] != label_filter:
continue
# Skip patterns without image data
if pattern["image_data"] is None:
continue
# Decode pattern image
pattern_image = self._decode_pattern_image(pattern)
if pattern_image is None:
continue
# Resize to match dimensions for comparison
resized_word, resized_pattern = self._resize_for_comparison(
gray_word, pattern_image
)
# Compute SSIM
score = self._compute_ssim(resized_word, resized_pattern, ssim_fn)
if score > best_score:
best_score = score
best_match = PatternMatch(
pattern_id=pattern["id"],
label=pattern["label"],
confidence=score,
pattern_image=resized_pattern,
)
# Check threshold
if best_match and best_match.confidence >= self.threshold:
try:
self.db.increment_pattern_use(best_match.pattern_id)
except Exception:
pass
logger.debug(
f"Pattern matched: '{best_match.label}' "
f"(score={best_match.confidence:.3f})"
)
return best_match
return None
def match_text_corrections(self, text: str) -> list[dict]:
"""Look up text-based corrections from the database.
Checks the corrections table for exact or similar text matches.
Args:
text: OCR text to look up corrections for.
Returns:
List of correction dictionaries with original, corrected,
and confidence.
"""
# Try exact match first
exact = self.db.find_correction(text)
if exact:
return [{
"original": exact["original_text"],
"corrected": exact["corrected_text"],
"confidence": 1.0,
"source": "exact",
}]
# Try corrections with similar original text
all_corrections = self.db.get_corrections(limit=100)
matches: list[dict] = []
for corr in all_corrections:
if corr["original_text"] == text:
matches.append({
"original": corr["original_text"],
"corrected": corr["corrected_text"],
"confidence": 1.0,
"source": "exact",
})
elif self._text_similarity(text, corr["original_text"]) > 0.8:
matches.append({
"original": corr["original_text"],
"corrected": corr["corrected_text"],
"confidence": self._text_similarity(
text, corr["original_text"]
),
"source": "fuzzy",
})
# Sort by confidence
matches.sort(key=lambda m: m["confidence"], reverse=True)
return matches
# ------------------------------------------------------------------
# Storing patterns
# ------------------------------------------------------------------
def store_word_pattern(
self,
label: str,
word_image: np.ndarray,
ocr_text: str = "",
confidence: float = 0.0,
source_engine: str = "",
) -> int:
"""Store a word pattern image in the database.
Used after user correction to save the corrected word image
as a pattern for future matching.
Args:
label: Correct text (user-provided).
word_image: Cropped word image to store.
ocr_text: Original OCR text.
confidence: Original OCR confidence.
source_engine: OCR engine that produced the result.
Returns:
Pattern ID in the database.
"""
# Convert to grayscale for storage efficiency
if len(word_image.shape) == 3:
gray = cv2.cvtColor(word_image, cv2.COLOR_BGR2GRAY)
else:
gray = word_image
# Resize to a standard size for consistency
target_height = 40
scale = target_height / max(gray.shape[0], 1)
target_width = max(int(gray.shape[1] * scale), 10)
resized = cv2.resize(gray, (target_width, target_height))
# Encode as PNG
success, encoded = cv2.imencode(".png", resized)
if not success:
logger.error("Failed to encode pattern image")
return -1
image_bytes = encoded.tobytes()
return self.db.add_pattern(
label=label,
image_data=image_bytes,
image_width=target_width,
image_height=target_height,
ocr_text=ocr_text,
confidence=confidence,
source_engine=source_engine,
)
# ------------------------------------------------------------------
# Statistics
# ------------------------------------------------------------------
def get_statistics(self) -> dict:
"""Get pattern matching statistics.
Returns:
Dictionary with pattern count, correction count, and other stats.
"""
stats = self.db.get_all_stats()
stats["cached_patterns"] = len(self._patterns_cache)
stats["threshold"] = self.threshold
return stats
# ------------------------------------------------------------------
# Internal helpers
# ------------------------------------------------------------------
@staticmethod
def _decode_pattern_image(pattern: dict) -> Optional[np.ndarray]:
"""Decode a pattern image from stored bytes.
Args:
pattern: Pattern dictionary with image_data field.
Returns:
Decoded grayscale image, or None if decoding fails.
"""
try:
if pattern["image_data"] is None:
return None
np_array = np.frombuffer(pattern["image_data"], dtype=np.uint8)
image = cv2.imdecode(np_array, cv2.IMREAD_GRAYSCALE)
if image is None:
return None
return image
except Exception as exc:
logger.debug(f"Failed to decode pattern image: {exc}")
return None
@staticmethod
def _resize_for_comparison(
img1: np.ndarray,
img2: np.ndarray,
) -> tuple[np.ndarray, np.ndarray]:
"""Resize two images to the same dimensions for SSIM comparison.
Resizes the smaller image to match the larger one's dimensions.
Args:
img1: First grayscale image.
img2: Second grayscale image.
Returns:
Tuple of (resized_img1, resized_img2) with matching dimensions.
"""
h1, w1 = img1.shape
h2, w2 = img2.shape
target_h = max(h1, h2)
target_w = max(w1, w2)
if h1 != target_h or w1 != target_w:
img1 = cv2.resize(img1, (target_w, target_h))
if h2 != target_h or w2 != target_w:
img2 = cv2.resize(img2, (target_w, target_h))
return img1, img2
@staticmethod
def _compute_ssim(
img1: np.ndarray,
img2: np.ndarray,
ssim_fn,
) -> float:
"""Compute SSIM between two images.
Args:
img1: First grayscale image.
img2: Second grayscale image.
ssim_fn: The structural_similarity function reference.
Returns:
SSIM score between -1.0 and 1.0.
"""
try:
if img1.shape != img2.shape:
return 0.0
score = ssim_fn(img1, img2, data_range=255)
return float(score)
except Exception as exc:
logger.debug(f"SSIM computation failed: {exc}")
return 0.0
@staticmethod
def _text_similarity(text1: str, text2: str) -> float:
"""Compute simple character-level similarity between two strings.
Uses character-level Jaccard similarity.
Args:
text1: First text string.
text2: Second text string.
Returns:
Similarity ratio between 0.0 and 1.0.
"""
if not text1 or not text2:
return 0.0
set1 = set(text1)
set2 = set(text2)
intersection = len(set1 & set2)
union = len(set1 | set2)
if union == 0:
return 1.0
return intersection / union
|