Skip to main content
NUTRITION 2023 Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Session: Mathematical and Measurement Advances to Enhance Nutritional Assessment (Poster Theater Flash Session 6)

(PTFS06-05-23) Portion Size Estimation of Food in a Bowl From Images

Sunday, July 23, 2023
4:21 PM – 4:25 PM ET
Location:  Fairfax
Disclosure(s):
Wenyan Jia, Dr.: No relevant financial relationship(s) with ineligible companies to disclose.
Megan A. McCrory, Dr.: Avocado Nutrition Science Advisory: Advisory Committee/Board Member (Ongoing)
Objectives: Image-based dietary assessment is important as it provides objective and quantitative food intake information. Portion size estimation from a food image is challenging when a scale reference is absent from the image (similar to a map without a scale bar). Objects with known sizes (e.g., a checkerboard card) have been used as scale references. However, for a bowl, providing a scale reference is inadequate because the food portion is also affected by the shape of the bowl. Instead of using a scale reference, our team has developed a new method to pre-calibrate the bowl using an imaging method prior to a dietary study. The surface of the bowl is first adhered with a paper ruler. Then, an image is taken from which the bowl is reconstructed using an algorithm. Our method and an interactive interface to help user visually estimate food portion have been reported and utilized in dietary studies. However, the bowl image must be taken directly from the top of the bowl. In practice, this procedure is difficult to perform accurately. In this work, we modify the computing method so that aiming the camera directly from the top of the bowl and visually estimating food portion are both eliminated.

Methods: Our computing method includes three steps. 1) Bowl reconstruction: We use an advanced optimization method to reconstruct the bowl in three-dimensional from the bowl image adhered with a paper ruler. 2) Bowl fullness estimation: A quantity called “bowl fullness” is defined as the ratio between the food volume and the volume of the bowl. The food area and the top boundary of the bowl in the image containing amorphous food (obtained from a dietary study) are identified and their ratio is calculated as the Food Area Ratio (FAR). After the relationship between the fullness and FAR is determined, the fullness is estimated. 3) Food portion estimation: The food portion is estimated by multiplying the fullness value and the estimated volume of the bowl from Step 1.

Results: Our experiment shows that, for 135 images of six different types of foods in a bowl with different portions, the mean absolute percentage error of our estimation was around 20%.

Conclusions: We have developed a practical image-based method to pre-calibrate a bowl and estimate the volume of an amorphous food in the bowl.

Funding Sources: This work was supported by the NIH and the Bill & Melinda Gates Foundation.