An Enhanced Pignistic Transformation-Based Fusion Scheme With Applications in Image Segmentation

Abstract

The traditional Pignistic transformation is limited in the context of “betting”, which faces information loss and is inconvenient for multi-source information fusion. To tackle this challenge, an Enhanced Pignistic transformation is proposed for the first time. New divergence and information volume measures are tailor-made for the enhanced Pignistic probability, and a novel information fusion algorithm is developed. To further prove the fusion algorithm’s advantages in conflict management, it is applied in a new semi-automatic image segmentation scheme. Two uncertain decision-support techniques named adaptive belief assignment and scalable information extraction are raised, and a fuzzy heuristic refinement algorithm is conducted, fulfilling the gap between evidential decision-making and segmentation refinement. Experimental analysis shows the proposed segmentation algorithm is superior on four metrics each and can enhance the robustness of foreground segmentation, indicating the effectiveness of the proposal in solving the decision inaccuracy of evidential segmentation schemes.

Keywords

• Enhanced Pignistic transformation
• multi-source information fusion
• belief hellinger distance
• semi-automatic image segmentation
• uncertain decision-support
• heuristic refinement