Chinese Dish Segmentation Based on Local Variation Driven Superpixel Grouping and Region Analysis

摘要

Food segmentation is a vital step to improve the accuracy of food recognition, which is important for automatic pricing and nutritional analysis of canteen. In this paper, we propose a segmentation algorithm based on local variation driven superpixel grouping and region analysis. This algorithm first uses superpixels to construct a graph model for Chinese dish images, and then uses the idea of local variation method to group superpixels with similar characteristics to form different regions. Finally, region analysis is used to extract regions belonging to dishes. Experimental results show that the proposed method can robustly segment Chinese dish images and extract dishes from images.