[Paper Review] A Review of the Vision-based Approaches for Dietary Assessment.

Ghalib Ahmed Tahir, Chu Kiong Loo|arXiv (Cornell University)|Jun 21, 2021

Nutritional Studies and DietMedicine89 references2 citations

TL;DR

This review evaluates vision-based approaches for automated dietary assessment, focusing on computer vision techniques for food recognition using image databases and mobile applications. It synthesizes state-of-the-art methodologies, evaluates their performance, and identifies research gaps and future challenges in mHealth-enabled dietary monitoring.

ABSTRACT

Last ten years have witnessed the growth of many computer vision applications for food recognition. Dietary studies showed that dietary-related problem such as obesity is associated with other chronic diseases like hypertension, irregular blood sugar levels, and increased risk of heart attacks. The primary cause of these problems is poor lifestyle choices and unhealthy dietary habits, which are manageable by using interactive mHealth apps that use automatic visual-based methods to assess dietary intake. This review discusses the most performing methodologies that have been developed so far for automatic food recognition. First, we will present the rationale of visual-based methods for food recognition. The core of the paper is the presentation, discussion and evaluation of these methods on popular food image databases. We also discussed the mobile applications that are implementing these methods. The review ends with a discussion of research gaps and future challenges in this area.

Motivation & Objective

• To examine the rationale and development of vision-based methods for automatic food recognition in dietary assessment.
• To evaluate the performance of leading computer vision techniques on standardized food image databases.
• To analyze existing mobile applications integrating these visual-based food recognition systems.
• To identify critical research gaps and future challenges in the field of automated dietary monitoring using computer vision.

Proposed method

• Systematic review of peer-reviewed literature on vision-based food recognition from the past decade.
• Categorization and comparative analysis of methodologies based on image processing, feature extraction, and classification techniques.
• Evaluation of methods using popular public food image databases such as Food-101, UEC-Food100, and others.
• Assessment of mobile applications implementing these vision-based systems for real-world dietary monitoring.
• Discussion of technical and practical challenges in deploying these systems in mHealth contexts.
• Synthesis of trends, limitations, and future research directions based on methodological and application-level findings.

Experimental results

Research questions

• RQ1 What are the most effective computer vision techniques for automatic food recognition in dietary assessment?
• RQ2 How do these vision-based methods perform across standardized food image databases?
• RQ3 Which mobile applications currently implement vision-based food recognition, and what are their technical and usability limitations?
• RQ4 What are the key research gaps and challenges in advancing vision-based dietary assessment systems?
• RQ5 How can vision-based approaches be further optimized for real-world mHealth applications?

Key findings

• Vision-based approaches have significantly advanced food recognition accuracy through deep learning and improved feature extraction techniques.
• Popular food image databases like Food-101 and UEC-Food100 have enabled standardized benchmarking of recognition models.
• Mobile applications leveraging these methods show promise in real-time dietary logging but face challenges in accuracy under real-world lighting and pose variations.
• Despite progress, significant performance gaps remain in recognizing mixed dishes, portion estimation, and low-light conditions.
• Current systems often lack generalization across diverse cuisines and food preparation styles.
• Future research must focus on improving robustness, interpretability, and integration into scalable mHealth platforms.