Authors : Hamirul Binti Hambali, Salwa Khalid Abdulateef, Massudi Mahmuddin, Nor Hazlyna Harun, Fadzilah Siraj and Hazaruddin Harun
Abstract: Image segmentation of food items is the most important stage in developing automated calorie estimation system. This stage refers to a process that classifies images into distinct regions with the aim to extract only the food image from the background. Currently, there are several segmentation methods which have been used in object identification. Active contour is one of the highly reputable method for image segmentation. However, this method suffers from few limitations in real world applications. Therefore, the goal of this study is to present a hybrid segmentation of multiple food items based on integrating active contour with K-means. After a theoretical reviewed of active contour, K-means is presented. Next, the segmentation process is presented starting from building the dataset, automating the contour initialization, conducting the homogenous test to determine the number of items in the region. The results show that, active contour has performed well for segmenting multiple food items when they were separated. However, it suffers from low capability in segmenting connected food items due to the similarity in visual metrics. Incorporating K-means with homogeneity test has shown good performance for segmentation of connected food items.
Hamirul Binti Hambali, Salwa Khalid Abdulateef, Massudi Mahmuddin, Nor Hazlyna Harun, Fadzilah Siraj and Hazaruddin Harun, 2018. Segmentation of Multi Food Images Using Integrated Active Contour and K-means. Journal of Engineering and Applied Sciences, 13: 3146-3151.