Food Recognition Based on Image Retrieval with Global Feature Descriptor

Abstract

This paper proposes a simple and effective non-parametric approach to solve the problem of food images parsing and label images with their categories. Firstly, the proposed approach works by six types of global image features: CEDD, FCTH, BTDH, EHD, CLD and SCD to matching with global image descriptors, labeling image with their categories, and the distance for each descriptor are fused to get the likelihood probability of each class, then efficient Markov random field (MRF) optimization is proposed for incorporating neighborhood context, besides optimization minimization are used Iterated Conditional Modes (ICM) algorithms. And this paper also introduces a non-parametric, data-driven approaches framework. This approach requires no training, just prior distribution and joint distribution are taken into account, and it can easily scale to data sets with tens of thousands of images and hundreds of labels. At last, the experiments show that the proposed method is significantly more accurate and faster at identifying food than existing methods.