| """Image preprocessing to improve OCR accuracy."""
|
|
|
| from __future__ import annotations
|
|
|
| import numpy as np
|
| from PIL import Image, ImageEnhance, ImageOps
|
|
|
|
|
| def preprocess_for_ocr(image: Image.Image, max_width: int = 1600) -> Image.Image:
|
| """Enhance contrast and resize for better OCR on scanned invoices."""
|
| img = image.convert("RGB")
|
|
|
| if img.width > max_width:
|
| ratio = max_width / img.width
|
| img = img.resize((max_width, int(img.height * ratio)), Image.Resampling.LANCZOS)
|
|
|
| gray = ImageOps.grayscale(img)
|
| enhanced = ImageEnhance.Contrast(gray).enhance(1.6)
|
| sharpened = ImageEnhance.Sharpness(enhanced).enhance(1.3)
|
| return sharpened.convert("RGB")
|
|
|
|
|
| def deskew_estimate(image: Image.Image) -> Image.Image:
|
| """Light deskew using numpy — skip heavy CV deps."""
|
| arr = np.array(image.convert("L"))
|
| if arr.size == 0:
|
| return image
|
|
|
| p5, p95 = np.percentile(arr, [5, 95])
|
| if p95 > p5:
|
| arr = np.clip((arr - p5) * 255.0 / (p95 - p5), 0, 255).astype(np.uint8)
|
| return Image.fromarray(arr).convert("RGB")
|
|
|
