Scalable image segmentation via decoupled sub-graph compression. (June 2018)

• Record Type: Journal Article
• Authors: Medeiros, R.S.; Wong, A.; Scharcanski, J.
• Is Part Of: Pattern recognition. Volume 78(2018:Jun.)
• Page Start: 228
• Page End: 241
• Publication Date: 2018-06
• Subjects: Segmentation -- Graph compression -- Decoupling -- Scalability; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.11.028
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A scalable graph compression algorithm for image segmentation proposed. The input image is represented by a region graph model. Texton dictionaries capture the local texture features in decoupled sub-graphs. A graph compression algorithm reduces the graph size and segments the image. Local graph decoupling and recoupling operations lead to an efficient method. Abstract: Dealing with large images is an on-going challenge in image segmentation, where many of the current methods run into computational and/or memory complexity issues. This work presents a novel decoupled sub-graph compression (DSC) approach for efficient and scalable image segmentation. In DSC, the image is modeled as a region graph, which is then decoupled into small sub-graphs. The sub-graphs undergo a compression process, which simplifies the graph, reducing the number of vertices and edges, while keeping the overall graph structure. Finally, the compressed sub-graphs are re-coupled and re-compressed to form a final compressed graph representing the final image segmentation. Experimental results based on a dataset of high resolution images (1000 × 1500) show that the DSC method achieves better segmentation performance when compared to state-of-the-art segmentation methods (PRI=0.84 and F=0.61), while having significantly lower computational and memory complexity.