A Comparison of Image Segmentation Algorithms

摘要: Unsupervised image segmentation algorithms have matured to the point where they generate reasonable segmentations, and thus can begin be incorporated into larger systems. A system designer now has an array of available algorithm choices, however, few objective numerical evaluations exist these algorithms. As a first step towards filling this gap, paper presents evaluation two popular algorithms, mean shift-based graph-based scheme. We also consider hybrid method which combines other methods. This quantitative is made possible by recently proposed measure correctness, Normalized Probabilistic Rand (NPR) index, allows principled comparison between segmentations created different as well on images. For each algorithm, we its correctness measured NPR stability with respect changes in parameter settings An produces correct results wide parameters any one image, multiple images same parameters, will useful, predictable easily adjustable preprocessing system. Our are presented Berkeley database, contains 300 natural along several ground truth hand for image. opposed previous compare all use features (position colour) segmentation, thereby making their outputs directly comparable.