Obesity is a leading cause of preventable death and disability in the U.S. Self- monitoring of all foods and beverages consumed is central to weight loss and maintenance efforts; however, this places a heavy burden on the user. These same burdens also impede nutritional research. The proposed research is for the testing of a semi-automated, objective, near real-time computer vision and pattern recognition approach to the measurement of dietary intake. In the proposed product, cell phone pictures of meals and snacks will be analyzed by software in an attempt to automatically recognize as many items as possible. A small number of intelligent yes/no questions will help provide additional information when necessary in order to meet the accuracy demands of the target application. Following identification of the items, the software will estimate the portion sizes of all identified items. The experiments comprising this Phase I SBIR are (a) extract the most informative sets of features using a large number of food and beverage items taken from an existing database of real world meal images, (b) compare the accuracy of candidate pattern recognition approaches to identify items based on the extracted features, (c) identify the most feasible algorithms for estimating portion size, and (d) test usability and user acceptance with a simulated version of the product. Phase II will (a) apply the approach to a greater variety of food and beverage items, (b) improve automated analysis, and (c) compare the approach to existing assessment instruments. This research will extend defense- and security-related technologies to the assessment and treatment of obesity. ? ? ?
