Paddle Ocr not detecting text when too much of the image is empty

Question

I got a multitude of envelopes captured like in this screenshot (address hidden by me for confidentiality). For most of them Paddle OCR is working great. But for a couples it doesn't captures much. If i crop the "empty" spaces and kept only the envelope it works fine. I got thousands of images so i can't crop them individually. Any clue why Paddle OCR is acting like that for some images and not for other that are similar? Any clue to correct that behavior?

result is giving me solely from the top-right part of the envelope : "Lettermall CON nuqme'

ocr = PaddleOCR(use_angle_cls=True, lang='en')
result = ocr.ocr(image_path, cls=True)