690530:1420 ADR-030-231-ocr-sandbox-two-step-flow #05.1 [skip ci]

This commit is contained in:
2026-05-30 14:20:54 +07:00
parent 30d9d721fb
commit b7a7b1e84d
@@ -33,8 +33,25 @@ app = FastAPI(title="Tesseract OCR Sidecar", version="1.0.0")
OCR_CHAR_THRESHOLD = int(os.getenv("OCR_CHAR_THRESHOLD", "100")) OCR_CHAR_THRESHOLD = int(os.getenv("OCR_CHAR_THRESHOLD", "100"))
MAX_PAGES = int(os.getenv("OCR_MAX_PAGES", "0")) # 0 = ทุกหน้า MAX_PAGES = int(os.getenv("OCR_MAX_PAGES", "0")) # 0 = ทุกหน้า
OCR_LANG = os.getenv("OCR_LANG", "tha+eng") # Tesseract language code (tha+eng = Thai + English) OCR_LANG = os.getenv("OCR_LANG", "tha+eng") # Tesseract language code (tha+eng = Thai + English)
# PSM 6 = Assume single uniform block of text (เหมาะกับเอกสารที่มี header/footer)
# OEM 1 = LSTM only (ดีกว่า legacy engine)
TESSERACT_CONFIG = f"--psm 6 --oem 1"
# Crop margin: ตัด header/footer (บน 10%, ล่าง 10%)
CROP_TOP_RATIO = 0.10
CROP_BOTTOM_RATIO = 0.02
logger.info(f"Tesseract OCR Sidecar initialized (lang={OCR_LANG})") logger.info(f"Tesseract OCR Sidecar initialized (lang={OCR_LANG}, config={TESSERACT_CONFIG})")
def crop_header_footer(pil_image: Image.Image, top_ratio: float = 0.10, bottom_ratio: float = 0.10) -> Image.Image:
"""Crop header/footer ออกจาก image เพื่อลบข้อความที่ไม่จำเป็น"""
width, height = pil_image.size
top_crop = int(height * top_ratio)
bottom_crop = int(height * bottom_ratio)
# Crop: (left, top, right, bottom)
cropped = pil_image.crop((0, top_crop, width, height - bottom_crop))
return cropped
def preprocess_image(pil_image: Image.Image) -> Image.Image: def preprocess_image(pil_image: Image.Image) -> Image.Image:
@@ -116,10 +133,14 @@ def ocr_extract(req: OcrRequest):
img_bytes = pix.tobytes("png") img_bytes = pix.tobytes("png")
img = Image.open(io.BytesIO(img_bytes)) img = Image.open(io.BytesIO(img_bytes))
# Preprocess ด้วย OpenCV เพื่อเพิ่มความแม่นยำ # Crop header/footer ก่อเพื่อลบข้อความที่ไม่จำเป็น
processed_img = preprocess_image(img) cropped_img = crop_header_footer(img, CROP_TOP_RATIO, CROP_BOTTOM_RATIO)
text = pytesseract.image_to_string(processed_img, lang=OCR_LANG) # Preprocess ด้วย OpenCV เพื่อเพิ่มความแม่นยำ
processed_img = preprocess_image(cropped_img)
# OCR ด้วย Tesseract โดยใช้ PSM 6 และ OEM 1
text = pytesseract.image_to_string(processed_img, lang=OCR_LANG, config=TESSERACT_CONFIG)
ocr_text_parts.append(text.strip()) ocr_text_parts.append(text.strip())
ocr_text = "\n".join(ocr_text_parts).strip() ocr_text = "\n".join(ocr_text_parts).strip()