Food recognition using Codebook-based model with sparse-coding

Recently, with the increasing of unhealthy diets and the attracted attention for healthy life, how to manage the dietary life is becoming more and more important. In this paper, we aim to construct a food-log system, which can auto-recognize the menu contents from food image taken by mobile phone. In order to increase recognition rate of food image, this research explores the typical codebook-based mode-Bag-of Feature (BOF), and the improved sparse-coding (Sc) based one for image representation, which is one of the main factors for affecting on food-log system. Furthermore, instead of pooling all the coded local features globally, the Spatial Pyramid Matching (SPM) strategy is adopted to integrated them from various spatial scale sub-regions. The experimental results validate that the Sc-based codebook model can achieve much better performance than the typically Bag-of-Feature on dish recognition.