Deep K-median: a robust unsupervised image segmentation method

Image segmentation is one of the most important tasks in the field of image processing. Although deep learning-based methods have achieved great success in image segmentation tasks, real-world images often contain mixed noise and lack segmentation labels. For such challenging problems, the segmentation accuracy of existing deep learning-based methods is significantly affected. We introduce the concept of depth K-median for the first time and propose an unsupervised image segmentation method based on depth K-median. This method integrates learning representations and clustering into a unified model while designing a novel unsupervised image segmentation algorithm. This algorithm alternately updates the parameters of the deep network and the clustering parameters. In an unsupervised setting, the new method not only effectively addresses the interference of mixed noise in image segmentation but also better learns the essential features of the data. To validate the effectiveness of the proposed method, extensive experiments on the general VOC 2012 and BSDS300 datasets, comparing existing deep learning-based methods and traditional machine learning methods, demonstrate the effectiveness and advantages of the new method.