ABSTRACT
Visual food recognition has gained popularity in recent years due
to its key roles in monitoring an individual’s diet and public health
management. The aim of mobile visual food recognition is to
recognize food dishes from pictures taken using portable devices
such as mobile phones/tablets, and then analyze and monitor the
dietary intakes of the users. Existing solutions for visual food
recognition employ Deep Neural Networks (DNNs) that are trained
as dish classification engines. These DNNs are typically deployed
in back-end servers. When an individual snaps the photo of a food
dish, the image is sent to this DNN-based classifier for recognition
and analysis. The results from this analysis are sent back to the
mobile end-user and logged in the end-user’s food journal for
monitoring and health management. This chapter gives an overview
of recent advances in visual food recognition systems and the
state-of-the-art solutions for dietary logging and management. We
explore various challenges in designing such systems and highlight
some solutions that address these challenges. In addition, we
introduce a Personalized Compact Network that incorporates user-
specific dietary habits/preferences into the DNNs and achieves
high recognition accuracy with low memory and energy foot
prints.
