An intelligent framework for food label processing and ingredient recognition

This study presents IntellectSnacks, an integrated framework for food-label analysis designed to assist users in identifying ingredients, additives, and potential allergens from packaged-food labels using a smartphone-based workflow. The framework combines image preprocessing, optical character recognition (OCR), text normalization, approximate string matching using Levenshtein distance, token-based entity processing, and supplementary external information retrieval. The revised evaluation used 983 recoverable food-label images and a fixed 50-image subset for controlled comparison of RapidOCR and Tesseract with original and Otsu-preprocessed images. Processing latency, output consistency, and sensitivity to controlled image-quality perturbations were evaluated. RapidOCR with Otsu preprocessing achieved the lowest median OCR processing time of 909.30 ms among the four evaluated configurations, with no OCR failures in the 50-image subset. Robustness analysis showed relatively stable output under moderate brightness changes, whereas mild blur and image downscaling caused greater reductions in output consistency. The revised evaluation did not recalculate OCR accuracy, CER, WER, ingredient precision, recall, or F1-score because independently verified reference annotations were unavailable. The findings provide reproducible evidence regarding the computational behavior and image-quality sensitivity of the evaluated pipeline. They also highlight the need for larger annotated datasets, independent ground-truth evaluation, device-level benchmarking, user-centered studies, and clinical validation before real-world effectiveness or health outcomes can be established.