Fresh food recognition using feature fusion

This paper presents a fresh food recognition system that utilizes the feature fusion extracted from food images captured from optical fibers embedded inside a chopping board. We exploit both local and global features including color, SURF and shape for image representation. In addition, we propose cost-based schemes for feature matching and the Borda count method for feature fusion. An experiment is conducted on our previous study's dataset, which consists of 1,800 images of 12 food ingredients for evaluating the proposed method. The results demonstrate that the overall recognition accuracies can be achieved 86% precision and 83% recall, which is significantly improved from our previous work on food recognition.