Spaces:
Sleeping
Sleeping
| import cv2 | |
| import numpy as np | |
| import logging | |
| from app.domain.schemas.core import ImageQualityMetrics | |
| from app.services.image_processing.geometry import estimate_skew_angle | |
| logger = logging.getLogger("ocr_microservice") | |
| def analyze_image_quality(image: np.ndarray) -> ImageQualityMetrics: | |
| """ | |
| Analyzes document image quality without blocking OCR execution. | |
| Computes dimensions, blur score, brightness, contrast, skew angle, and resolution health. | |
| """ | |
| if image is None or image.size == 0: | |
| return ImageQualityMetrics( | |
| width=0, | |
| height=0, | |
| channels=0, | |
| is_blurry=True, | |
| blur_score=0.0, | |
| brightness=0.0, | |
| contrast=0.0, | |
| resolution_ok=False, | |
| skew_angle=0.0, | |
| detected_orientation=0, | |
| variants_evaluated=[], | |
| warnings=["Empty or invalid image."] | |
| ) | |
| height, width = image.shape[:2] | |
| channels = image.shape[2] if len(image.shape) == 3 else 1 | |
| # Convert to grayscale if necessary | |
| if channels == 3: | |
| gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) | |
| else: | |
| gray = image | |
| warnings = [] | |
| # 1. Blur Detection via Laplacian Variance | |
| try: | |
| laplacian = cv2.Laplacian(gray, cv2.CV_64F) | |
| blur_score = float(laplacian.var()) | |
| except Exception as e: | |
| logger.warning(f"Failed to calculate blur score: {e}") | |
| blur_score = 100.0 | |
| # Blur threshold: typically < 80 indicates blur on document scans | |
| is_blurry = blur_score < 80.0 | |
| if is_blurry: | |
| warnings.append(f"Image may be blurry (variance: {blur_score:.1f}). Text recognition accuracy may be reduced.") | |
| # 2. Brightness and Contrast | |
| brightness = float(np.mean(gray)) | |
| contrast = float(np.std(gray)) | |
| if brightness < 40.0: | |
| warnings.append(f"Image appears underexposed (mean brightness: {brightness:.1f}/255).") | |
| elif brightness > 230.0: | |
| warnings.append(f"Image appears washed out or overexposed (mean brightness: {brightness:.1f}/255).") | |
| if contrast < 25.0: | |
| warnings.append(f"Low contrast detected ({contrast:.1f}). Text may blend with background.") | |
| # 3. Resolution Check | |
| min_dim = min(width, height) | |
| resolution_ok = min_dim >= 300 | |
| if not resolution_ok: | |
| warnings.append(f"Low resolution image ({width}x{height}). Recommended minimum dimension is 300px.") | |
| # Cardinal orientation is not inferred from aspect ratio: that cannot | |
| # distinguish 90 from 270 degrees or portrait content from a rotated scan. | |
| detected_orientation = 0 | |
| # 4. Skew Angle Estimation | |
| skew_angle = 0.0 | |
| try: | |
| skew_angle = estimate_skew_angle(gray) | |
| if abs(skew_angle) >= 1.0: | |
| warnings.append(f"Document skew detected ({skew_angle:+.1f}°). Automatic deskew recommended.") | |
| except Exception as e: | |
| logger.debug(f"Skew estimation skipped: {e}") | |
| return ImageQualityMetrics( | |
| width=width, | |
| height=height, | |
| channels=channels, | |
| is_blurry=is_blurry, | |
| blur_score=round(blur_score, 2), | |
| brightness=round(brightness, 2), | |
| contrast=round(contrast, 2), | |
| resolution_ok=resolution_ok, | |
| skew_angle=skew_angle, | |
| detected_orientation=detected_orientation, | |
| variants_evaluated=[], | |
| warnings=warnings | |
| ) | |