4D Facial shape measurements during mastication and deep learning-based analysisfor quantitative evaluation instead of sensory testing

~ A case study of flavor effect ~

Abstract

Most of palatability comes from food texture, especially in case of Japanese foods (>60%). Evaluation of texture, both by sensory and by instrumental tests, is quite difficult. Although sensory test is essentially unstable, it is most important at least for final evaluation. Instrumental test is stable, but the results do not agree with the sensory test probably due to large difference of experimental condition. We developed facial 3D shape measurement system as a function of time during mastication. Collecting data of this method is relatively easier than the other methods particularlly for big data for deep learning, another advantage is that facial motion during mastication reflects both human evaluation and physical properties of foods such as texture. As demonstration, this method was applied to detect the effect of flavor and/or taste on mastication behavior. It is found that this reflects food texture, such as hardness and the other texture characteristics, and taste, flavor effects, based on deep learning. The cross-correlation between flavor and taste can also be detected by this 4D facial motion measurement during mastication.