457546754.py

457546754.py

from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image

import pytesseract

# Convert the previously corrected image to RGB (required for pytesseract)
warped_rgb = cv2.cvtColor(warped, cv2.COLOR_BGR2RGB)

# Use pytesseract to extract text
extracted_text = pytesseract.image_to_string(warped_rgb, lang='ron')

# Return the text
extracted_text[:2000]  # Show first 2000 characters for preview


# Load the image again, resize for better OCR performance
ocr_image = warped_rgb
ocr_pil_image = Image.fromarray(ocr_image).resize((ocr_image.shape[1]//2, ocr_image.shape[0]//2))

# Load TrOCR processor and model (OCR with Transformers)
from transformers import TrOCRProcessor, VisionEncoderDecoderModel

processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")

# Preprocess image and generate prediction
pixel_values = processor(images=ocr_pil_image, return_tensors="pt").pixel_values
generated_ids = model.generate(pixel_values)
ocr_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

ocr_text