Spaces:
Sleeping
Sleeping
File size: 10,872 Bytes
5f3b8de | 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 | """
Image Preprocessor β Adaptive image analysis and preprocessing for OCR.
Analyzes image properties (resolution, contrast, noise, skew) and applies
optimal preprocessing pipeline. Part of the agentic system's perception layer.
"""
import cv2
import numpy as np
from PIL import Image
from dataclasses import dataclass, field
from typing import Optional, Tuple
import logging
logger = logging.getLogger(__name__)
@dataclass
class ImageProperties:
"""Properties extracted from image analysis."""
width: int = 0
height: int = 0
resolution_dpi: int = 72
contrast_score: float = 0.0 # 0-1 scale
brightness_score: float = 0.0 # 0-1 scale
noise_level: float = 0.0 # 0-1 scale (higher = noisier)
skew_angle: float = 0.0 # degrees
is_grayscale: bool = False
is_binary: bool = False
sharpness_score: float = 0.0 # 0-1 scale
quality_rating: str = "unknown" # "excellent", "good", "fair", "poor"
@dataclass
class PreprocessingConfig:
"""Configuration for preprocessing steps, determined by image analysis."""
apply_grayscale: bool = True
apply_clahe: bool = True
clahe_clip_limit: float = 2.0
clahe_grid_size: Tuple[int, int] = (8, 8)
apply_denoise: bool = True
denoise_strength: int = 10
apply_binarize: bool = True
binarize_method: str = "otsu" # "otsu", "adaptive", "none"
apply_deskew: bool = False
apply_sharpen: bool = False
apply_resize: bool = False
target_dpi: int = 300
def analyze_image(image_path: str) -> ImageProperties:
"""
Analyze an image and extract its properties for the agent to make
preprocessing decisions.
Args:
image_path: Path to the input image.
Returns:
ImageProperties with analysis results.
"""
props = ImageProperties()
img = cv2.imread(image_path)
if img is None:
raise FileNotFoundError(f"Cannot open image: {image_path}")
props.height, props.width = img.shape[:2]
props.is_grayscale = len(img.shape) == 2 or (len(img.shape) == 3 and img.shape[2] == 1)
# Convert to grayscale for analysis
if not props.is_grayscale:
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
else:
gray = img if len(img.shape) == 2 else img[:, :, 0]
# Contrast score (standard deviation of pixel values, normalized)
std_dev = np.std(gray.astype(np.float64))
props.contrast_score = min(std_dev / 80.0, 1.0)
# Brightness score (mean pixel value, normalized)
mean_val = np.mean(gray.astype(np.float64))
props.brightness_score = mean_val / 255.0
# Noise estimation (using Laplacian variance)
laplacian_var = cv2.Laplacian(gray, cv2.CV_64F).var()
props.sharpness_score = min(laplacian_var / 500.0, 1.0)
# High Laplacian can also mean noise; estimate noise from smooth regions
blur = cv2.GaussianBlur(gray, (5, 5), 0)
noise_estimate = np.mean(np.abs(gray.astype(np.float64) - blur.astype(np.float64)))
props.noise_level = min(noise_estimate / 30.0, 1.0)
# Check if already binary
unique_vals = len(np.unique(gray))
props.is_binary = unique_vals <= 10
# Skew detection via Hough Line Transform
props.skew_angle = _detect_skew(gray)
# DPI estimation from image metadata
try:
pil_img = Image.open(image_path)
dpi_info = pil_img.info.get('dpi', (72, 72))
props.resolution_dpi = int(dpi_info[0]) if isinstance(dpi_info, tuple) else int(dpi_info)
except Exception:
props.resolution_dpi = 72
# Overall quality rating
props.quality_rating = _rate_quality(props)
logger.info(f"Image analysis: {props.width}x{props.height}, "
f"contrast={props.contrast_score:.2f}, noise={props.noise_level:.2f}, "
f"quality={props.quality_rating}")
return props
def _detect_skew(gray: np.ndarray) -> float:
"""Detect text skew angle using Hough line transform."""
try:
edges = cv2.Canny(gray, 50, 150, apertureSize=3)
lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 100,
minLineLength=gray.shape[1] // 4,
maxLineGap=10)
if lines is None or len(lines) == 0:
return 0.0
angles = []
for line in lines:
x1, y1, x2, y2 = line[0]
angle = np.degrees(np.arctan2(y2 - y1, x2 - x1))
# Only consider near-horizontal lines
if abs(angle) < 15:
angles.append(angle)
if angles:
return float(np.median(angles))
except Exception as e:
logger.warning(f"Skew detection failed: {e}")
return 0.0
def _rate_quality(props: ImageProperties) -> str:
"""Rate overall image quality for OCR."""
score = 0
# Resolution
if props.resolution_dpi >= 300:
score += 3
elif props.resolution_dpi >= 150:
score += 2
else:
score += 1
# Contrast
if props.contrast_score > 0.6:
score += 3
elif props.contrast_score > 0.3:
score += 2
else:
score += 1
# Noise
if props.noise_level < 0.2:
score += 3
elif props.noise_level < 0.5:
score += 2
else:
score += 1
# Sharpness
if props.sharpness_score > 0.4:
score += 3
elif props.sharpness_score > 0.15:
score += 2
else:
score += 1
if score >= 10:
return "excellent"
elif score >= 7:
return "good"
elif score >= 5:
return "fair"
else:
return "poor"
def determine_preprocessing(props: ImageProperties) -> PreprocessingConfig:
"""
Agent decision function: determine optimal preprocessing based on
image properties. This is the intelligence layer for preprocessing.
Args:
props: Image properties from analysis.
Returns:
PreprocessingConfig with recommended preprocessing steps.
"""
config = PreprocessingConfig()
# Already binary? Skip binarization
if props.is_binary:
config.apply_binarize = False
config.apply_clahe = False
logger.info("Image already binary β skipping CLAHE and binarization")
# Low contrast β stronger CLAHE
if props.contrast_score < 0.3:
config.apply_clahe = True
config.clahe_clip_limit = 3.0
logger.info("Low contrast detected β increasing CLAHE clip limit")
elif props.contrast_score > 0.7:
config.apply_clahe = False
logger.info("High contrast β CLAHE not needed")
# High noise β stronger denoising
if props.noise_level > 0.5:
config.apply_denoise = True
config.denoise_strength = 15
logger.info("High noise β increasing denoise strength")
elif props.noise_level < 0.15:
config.apply_denoise = False
logger.info("Low noise β denoising not needed")
# Skewed β deskew
if abs(props.skew_angle) > 0.5:
config.apply_deskew = True
logger.info(f"Skew detected ({props.skew_angle:.1f}Β°) β enabling deskew")
# Low sharpness β sharpen
if props.sharpness_score < 0.15:
config.apply_sharpen = True
logger.info("Low sharpness β enabling sharpening")
# Low resolution β upscale
if props.resolution_dpi < 150 and max(props.width, props.height) < 1500:
config.apply_resize = True
logger.info("Low resolution β enabling upscaling")
# Adaptive binarization for uneven lighting
if props.brightness_score < 0.3 or props.brightness_score > 0.7:
config.binarize_method = "adaptive"
logger.info("Uneven brightness β using adaptive binarization")
return config
def preprocess_image(image_path: str, config: Optional[PreprocessingConfig] = None) -> np.ndarray:
"""
Preprocess an image for OCR based on the given configuration.
Args:
image_path: Path to the input image.
config: Preprocessing configuration. If None, auto-determine.
Returns:
Preprocessed image as numpy array.
"""
img = cv2.imread(image_path)
if img is None:
raise FileNotFoundError(f"Cannot open image: {image_path}")
if config is None:
props = analyze_image(image_path)
config = determine_preprocessing(props)
# Step 1: Grayscale
if config.apply_grayscale and len(img.shape) == 3:
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
elif len(img.shape) == 2:
gray = img
else:
gray = img[:, :, 0] if img.shape[2] == 1 else cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Step 2: Resize/upscale
if config.apply_resize:
scale = config.target_dpi / 72.0
scale = min(scale, 3.0) # Cap at 3x
new_w = int(gray.shape[1] * scale)
new_h = int(gray.shape[0] * scale)
gray = cv2.resize(gray, (new_w, new_h), interpolation=cv2.INTER_CUBIC)
logger.info(f"Resized to {new_w}x{new_h}")
# Step 3: Deskew
if config.apply_deskew:
gray = _deskew(gray)
# Step 4: CLAHE contrast enhancement
if config.apply_clahe:
clahe = cv2.createCLAHE(
clipLimit=config.clahe_clip_limit,
tileGridSize=config.clahe_grid_size
)
gray = clahe.apply(gray)
# Step 5: Denoise
if config.apply_denoise:
gray = cv2.fastNlMeansDenoising(gray, h=config.denoise_strength)
# Step 6: Sharpen
if config.apply_sharpen:
kernel = np.array([[-1, -1, -1],
[-1, 9, -1],
[-1, -1, -1]])
gray = cv2.filter2D(gray, -1, kernel)
# Step 7: Binarize
if config.apply_binarize:
if config.binarize_method == "otsu":
_, gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
elif config.binarize_method == "adaptive":
gray = cv2.adaptiveThreshold(
gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2
)
return gray
def _deskew(image: np.ndarray) -> np.ndarray:
"""Correct text skew in an image."""
coords = np.column_stack(np.where(image > 0))
if len(coords) < 100:
return image
try:
angle = cv2.minAreaRect(coords)[-1]
if angle < -45:
angle = -(90 + angle)
else:
angle = -angle
if abs(angle) < 0.5:
return image
h, w = image.shape[:2]
center = (w // 2, h // 2)
M = cv2.getRotationMatrix2D(center, angle, 1.0)
rotated = cv2.warpAffine(
image, M, (w, h),
flags=cv2.INTER_CUBIC,
borderMode=cv2.BORDER_REPLICATE
)
logger.info(f"Deskewed by {angle:.2f}Β°")
return rotated
except Exception as e:
logger.warning(f"Deskew failed: {e}")
return image
|