Food image recognition is different from general object recognition tasks. Many types of ingredients in the food images do not have a unique spatial layout, and the shape of ingredients may change with cooking methods and cutting methods.
This is a big challenge for food recognition, especially, recognizing the ingredients in the food images. In this paper, we focus on recognizing the ingredients segmented from the food images, and propose a method for it. In details, the candidate regions of the ingredients for the recognition are located by the methods of locating and sliding windows. Then, these regions are assigned to the ingredient classes by the CNN-based ingredients classifier trained by the dataset with single ingredient images. Finally, the final ingredients are decided from these candidate results by using the scheme of decision making.The effectiveness of the proposed method is evaluated by the experimental results.