Data clustering is an important machine-learning topic. It is useful for variety of applications one of them is image segmentation. A given divided image into regions homogenous additional to certain features is the image segmentation process, which matches real objects of an actual scene. FIS (Food Image Segmentation) is important for calories estimation. K-means has been used for performing such task. However, in order to conclude the food items number in the image, it requires interacting with the application. This article, presents a novel approach based dependently on k-means named Hk-means (Homogeneity test of k-means) is developed to calculate k value and applied for FIS for the purpose of assuring full autonomy in the calories estimation system. This approach uses the homogeneity test so as to compensate the new item existence in the image. The suggested method Hk-means is tested on food images and show accuracy 96%. The experimental results has achieved 1.5 second execution time when compare with benchmark method.