Unsupervised RGB-D image segmentation by multi-layer clustering

Abstract

Microsoft Kinect is an imaging sensor which provides access to synchronized depth and color image, also called RGB-D image. In this paper, we propose a clustering based unsupervised method for RGB-D image segmentation. To this aim, we follow a multi-layer clustering strategy. The first layer generates multiple segmentations using the K-Means clustering with different parameters and different types of feature. The second layer combines these results and generates the final segmentation using the segmentation by aggregating superpixels (SAS) method. We evaluate our method on the NYUD2 database and compare it with the state-of-the-art methods. Results show that our method is better than the existing clustering based RGB-D segmentation methods, comparable with the state-of-the-art and requires less computation time. Moreover, it provides numerous perspectives to extend and improve in future.