Date of Award:
Doctor of Philosophy (PhD)
An important component of a healthy diet is the comprehension and retention of nutritional information and understanding of how different food items and nutritional constituents affect our bodies. In the U.S. and many other countries, nutritional information is primarily conveyed to consumers through nutrition labels (NLs) which can be found in all packaged food products. However, sometimes it becomes really challenging to utilize all this information available in these NLs even for consumers who are health conscious as they might not be familiar with nutritional terms or find it difficult to integrate nutritional data collection into their daily activities due to lack of time, motivation, or training. So it is essential to automate this data collection and interpretation process by integrating Computer Vision based algorithms to extract nutritional information from NLs because it improves the user’s ability to engage in continuous nutritional data collection and analysis. To make nutritional data collection more manageable and enjoyable for the users, we present a Proactive NUTrition Management System (PNUTS). PNUTS seeks to shift current research and clinical practices in nutrition management toward persuasion, automated nutritional information processing, and context-sensitive nutrition decision support.
PNUTS consists of two modules, firstly a barcode scanning module which runs on smart phones and is capable of vision-based localization of One Dimensional (1D) Universal Product Code (UPC) and International Article Number (EAN) barcodes with relaxed pitch, roll, and yaw camera alignment constraints. The algorithm localizes barcodes in images by computing Dominant Orientations of Gradients (DOGs) of image segments and grouping smaller segments with similar DOGs into larger connected components. Connected components that pass given morphological criteria are marked as potential barcodes. The algorithm is implemented in a distributed, cloud-based system. The system’s front end is a smartphone application that runs on Android smartphones with Android 4.2 or higher. The system’s back end is deployed on a five node Linux cluster where images are processed. The algorithm was evaluated on a corpus of 7,545 images extracted from 506 videos of bags, bottles, boxes, and cans in a supermarket. The DOG algorithm was coupled to our in-place scanner for 1D UPC and EAN barcodes. The scanner receives from the DOG algorithm the rectangular planar dimensions of a connected component and the component’s dominant gradient orientation angle referred to as the skew angle. The scanner draws several scan lines at that skew angle within the component to recognize the barcode in place without any rotations. The scanner coupled to the localizer was tested on the same corpus of 7,545 images. Laboratory experiments indicate that the system can localize and scan barcodes of any orientation in the yaw plane, of up to 73.28 degrees in the pitch plane, and of up to 55.5 degrees in the roll plane. The videos have been made public for all interested research communities to replicate our findings or to use them in their own research. The front end Android application is available for free download at Google Play under the title of NutriGlass. This module is also coupled to a comprehensive NL database from which nutritional information can be retrieved on demand. Currently our NL database consists of more than 230,000 products.
The second module of PNUTS is an algorithm whose objective is to determine the text skew angle of an NL image without constraining the angle’s magnitude. The horizontal, vertical, and diagonal matrices of the (Two Dimensional) 2D Haar Wavelet Transform are used to identify 2D points with significant intensity changes. The set of points is bounded with a minimum area rectangle whose rotation angle is the text’s skew. The algorithm’s performance is compared with the performance of five text skew detection algorithms on 1001 U.S. nutrition label images and 2200 single- and multi-column document images in multiple languages. To ensure the reproducibility of the reported results, the source code of the algorithm and the image data have been made publicly available. If the skew angle is estimated correctly, optical character recognition (OCR) techniques can be used to extract nutrition information.
Zaman, Tanwir, "Vision Based Extraction of Nutrition Information from Skewed Nutrition Labels" (2016). All Graduate Theses and Dissertations. 4893.
Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .