THE USE OF CO-OCCURRENCE PATTERNS IN SINGLE IMAGE BASED FOOD PORTION ESTIMATION

Area of Science:

• Nutrition science
• Computer vision
• Health informatics

Background:

• Accurate dietary intake measurement is a significant challenge in nutrition and health research.
• Estimating food portion sizes from images is difficult due to variations in food preparation and appearance.

Purpose of the Study:

• To develop and evaluate a novel method for improving food portion estimation accuracy.
• To leverage co-occurrence patterns as contextual information for more precise dietary assessment.

Main Methods:

• A geometric model-based technique was employed for initial food portion estimation.
• Co-occurrence patterns of food items were derived from images collected using the mobile Food Record (mFR) system.
• These co-occurrence patterns were integrated as prior knowledge to refine the portion estimation results.

Main Results:

• The study demonstrated that incorporating food co-occurrence patterns significantly enhances portion estimation accuracy.
• The mobile Food Record (mFR) system facilitated the collection of relevant dietary image data.
• Geometric modeling combined with contextual co-occurrence information proved effective for portion size estimation.

Conclusions:

• Utilizing co-occurrence patterns as contextual information is a promising approach to improve food portion estimation.
• The developed methodology offers a potential solution to the long-standing problem of accurate dietary intake measurement.
• This research contributes to advancing digital health tools for nutrition monitoring and research.