Semantic Image Segmentation Based on Spatial Context Relations

摘要

In this paper, the semantic image segmentation framework based on spatial context relations is proposed. First, the knowledge representation scheme of an image is introduced, which include fuzzy ontology structure and spatial context relations. For the purpose of the initial labeling of segmented regions, the multiclass fuzzy Support Vector Machine (multi-FSVM) is employed. A new image segmentation algorithm with high-level semantics is proposed, which is based on spatial context relations. At last, the experimental results are illustrated, and the advantage of the proposed image segmentation method is discussed.