On Comparing Color Spaces for Food Segmentation

Accurate segmentation of food regions is important for both food recognition and quantity estimation and any error would degrade the accuracy of the food dietary assessment system. Main goal of this work is to investigate the performance of a number of color encoding schemes and color spaces for food segmentation exploiting the JSEG algorithm. Our main outcome is that significant improvements in segmentation can be achieved with a proper color space selection and by learning the proper setting of the segmentation parameters from a training set.