Automatic segmentation of images and videos is a very challenging task in computer vision and one of the most crucial steps toward image and video understanding. In this research work we propose to include semantic criteria in the segmentation process to capture the semantic properties of objects that visual features, such as color or texture, are not able to describe.