Recognition and Classification of Fast Food Images
Food recognition is of great importance nowadays for multiple purposes. On one hand, for people who want to get a better understanding of the food that they are not familiar of or they haven’t even seen before, they can simply take a picture and get to know more details about it. On the other hand, the increasing demand for dietary assessment tools to record the calorie and nutrition has also been a driving force in the development of food recognition technique. Therefore, automatic food recognition is very important and has great application potential. However, food varies greatly in appearance (e.g., shape, colors) with tons of different ingredients and assembling methods. This makes food recognition a difficult task for current state-of-the-art classification methods, and hence an important challenge for Computer Vision researchers. Yoshiyuki Kawano and Keiji Yanai [Kawano and Yanai, 2013] proposed a real-time food recognition system which adopted a linear SVM with a fast χ 2 kernel, bounding box adjustment and estimation of the expected direction of a food region. Lukas Bossard et al. [Bossard et al., 2014] presented a novel method based on Random Forests to mine discriminative visual components and efficient classification. In this effort, we intend to utilize learned machine learning algorithms to do fast food recognition and classification. Our goal is to develop a computationally efficient algorithm with high accuracy. Different features and models have been implemented and compared.
Research Paper Link: Download Paper