Image Processing for the Extraction of Nutritional Information from Food Labels
This repository contains code and documentation for the senior thesis by Rick Sullivan and Nate Matsunaga. This project won best in session for Computer Science & Engineering in June 2015 as determined by industry judges. The thesis can be found here.
Current techniques for tracking nutritional data require undesirable amounts of either time or man- power. People must choose between tediously recording and updating dietary information or de- pending on unreliable crowd-sourced or costly maintained databases. Our project looks to overcome these pitfalls by providing a programming interface for image analysis that will read and report the information present on a nutrition label directly. Our solution provides a library that combines image pre-processing, optical character recog- nition, and post-processing techniques to pull the relevant information from an image of a nutrition label. We apply an understanding of a nutrition label’s content and data organization to approach the accuracy of traditional data-entry methods. Our system currently provides around 80% accuracy for most label images, and we will continue to work to improve our accuracy.