An Improved Depth Image Clustering Algorithm Based on Spatial Information

Depth image spatial clustering is an important task in the fields of computer vision and machine learning, aiming to group pixels or point cloud data of depth images into clusters with similar features. This is crucial for tasks such as object recognition, scene segmentation, 3D reconstruction, and other applications. The K-means algorithm is a commonly used clustering method that divides data points into K clusters, ensuring that each data point belongs to the cluster whose center is closest to it. However, traditional K-means algorithms face several challenges in depth image spatial clustering, such as sensitivity to the choice of K and the influence of cluster center initialization on the results. In this study, we propose an improved depth image clustering method based on spatial information. The method includes depth image edge extraction, intra-edge region growing, unclassified point re-clustering, and connected region merging. Through experiments conducted on datasets from TUM, Bonn, and our collected data, our algorithm demonstrates superior performance in depth image clustering tasks, higher clustering accuracy, fewer unclassified points, and faster convergence speed.