Combining multiple visual cognition systems for joint decision-making using combinatorial fusion

Abstract

Cognitive decision-making based on visual sensory input has been a topic of intensive interest. When combining these visual cognition systems, three models of strategy have been proposed and widely used: M₁: simple average, M₂: weighted average using σ, and M₃: weighted average using σ². In this paper, we extend each visual cognition system to a scoring system using combinatorial fusion analysis, a framework which has proven effective for the optimization of multiple evaluation methods in several other computational domains. Ten experiments on two visual systems each are conducted. M₁, M₂, M₃, and combinatorial fusion on M₁ are computed. Our two main results are: (a) Six of the ten experiments show performance order M₁ >; M₂ >; M₃ while the other four experiments exhibit the opposite order; (b) Combinatorial fusion based on M₁ performs better than M₁ in eight of the ten experiments. It is demonstrated that Combinatorial Fusion Analysis is useful in the study of visual cognition. Our results exhibit a new method to better analyze and make joint decisions in visual cognition using Combinatorial Fusion Analysis.