c74539b04b
New confidence.py module with two functions: - apply_confidence_markers(): replaces words below threshold with [unleserlich], collapses adjacent markers into one - words_from_characters(): reconstructs word-level confidence from Kraken's character-level data Surya 0.17 provides native word-level confidence via line.words. Kraken 7.0 provides per-character confidences via record.confidences. Both engines now pass word+confidence data through main.py, which applies the marker post-processing before returning the API response. Threshold configurable via OCR_CONFIDENCE_THRESHOLD env var (default 0.3). Frontend already renders [unleserlich] markers via transcriptionMarkers.ts. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
77 lines
2.6 KiB
Python
77 lines
2.6 KiB
Python
"""Surya OCR engine wrapper — transformer-based, handles typewritten and modern Latin handwriting."""
|
|
|
|
import logging
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_recognition_predictor = None
|
|
_detection_predictor = None
|
|
|
|
|
|
def load_models():
|
|
"""Eagerly load Surya models into memory. Called once at container startup."""
|
|
global _recognition_predictor, _detection_predictor
|
|
|
|
logger.info("Loading Surya models...")
|
|
|
|
from surya.foundation import FoundationPredictor
|
|
from surya.recognition import RecognitionPredictor
|
|
from surya.detection import DetectionPredictor
|
|
|
|
foundation_predictor = FoundationPredictor()
|
|
_recognition_predictor = RecognitionPredictor(foundation_predictor)
|
|
_detection_predictor = DetectionPredictor()
|
|
|
|
logger.info("Surya models loaded successfully")
|
|
|
|
|
|
def extract_blocks(images: list, language: str = "de") -> list[dict]:
|
|
"""Run Surya OCR on a list of PIL images (one per page).
|
|
|
|
Returns a flat list of block dicts with pageNumber, x, y, width, height,
|
|
polygon, text. Coordinates are normalized to [0, 1] relative to page dimensions.
|
|
Surya 0.17+ returns polygon (4-point) natively on each text line.
|
|
"""
|
|
all_blocks = []
|
|
|
|
predictions = _recognition_predictor(images, det_predictor=_detection_predictor)
|
|
|
|
for page_idx, page_pred in enumerate(predictions):
|
|
page_w, page_h = images[page_idx].size
|
|
|
|
for line in page_pred.text_lines:
|
|
bbox = line.bbox # [x1, y1, x2, y2] in pixel coordinates
|
|
x1, y1, x2, y2 = bbox
|
|
|
|
# Surya 0.17 provides polygon as list of (x, y) tuples (4 points, clockwise)
|
|
polygon = None
|
|
if hasattr(line, "polygon") and line.polygon and len(line.polygon) == 4:
|
|
polygon = [
|
|
[p[0] / page_w, p[1] / page_h]
|
|
for p in line.polygon
|
|
]
|
|
|
|
# Extract word-level confidence for [unleserlich] marking
|
|
words = []
|
|
if hasattr(line, "words") and line.words:
|
|
for word in line.words:
|
|
words.append({
|
|
"text": word.text,
|
|
"confidence": word.confidence,
|
|
})
|
|
else:
|
|
words = [{"text": line.text, "confidence": getattr(line, "confidence", 1.0)}]
|
|
|
|
all_blocks.append({
|
|
"pageNumber": page_idx,
|
|
"x": x1 / page_w,
|
|
"y": y1 / page_h,
|
|
"width": (x2 - x1) / page_w,
|
|
"height": (y2 - y1) / page_h,
|
|
"polygon": polygon,
|
|
"text": line.text,
|
|
"words": words,
|
|
})
|
|
|
|
return all_blocks
|