Packaging Label Design Automated Proofreading Review

Problem It Solves

Label reviews often require comparing text inside a design image with fields, dimensions, barcodes, and regulatory requirements in a specification document. Manual checks can miss small fields, misread dense nutrition tables, or incorrectly treat formatting differences such as spaces in GB 19295 or SC 10334012231076.

How It Works

The skill expects both a label image and requirement input. It parses the required label fields, then runs scripts/enhance_label_image.py to preprocess the image: high-resolution images are read at 1:1, smaller images are upscaled 3x, barcodes are decoded with pyzbar, RapidOCR provides a second text pass, and character-gap analysis corrects spaces lost by OCR. It then checks a mandatory 18-field checklist, including product name, standard code, ingredient list, net content, production date, shelf life, storage conditions, and nutrition facts, marking each item as passed, failed, or missing. Packaging checks cover dimensions, label format, typography, visual design, and carton labeling.

Boundaries

Images cannot directly verify physical print size, so required dimensions must be confirmed by the user. If the image is blurry, barcode detection fails, or OCR conflicts with visual reading, the skill flags manual review rather than forcing a conclusion. Regulatory checks use GB 7718 and GB 28050 as supplementary references, with the requirement document taking priority.