Image segmentation using morphological tools

Abstract

In this work we study the extraction of semantic objects from images together with a metric that ranks them according to their perceptual significance. To obtain an initial segmentation we use elements of mathematical morphology (level sets and level lines) and some properties such as T-junctions, contrast and compactness. Then, to refine the initial partition we apply regularization techniques and a standard merging algorithm. Finally, we compute a perceptual metric using factors that influence our perception.