This paper presents the results of an investigation into objective evaluation of natural colour image fusion. Preserving true colour information is vital for natural appearance of fused images and measures of fusion success should take this into account either within their overall fusion performance or in explicit naturalness measures. We adopt the well known gradient based objective fusion evaluation framework that evaluates preservation of input gradients in the fused images to focus our investigation onto metric configurations and representation spaces for comparing true colour information between colour input and fused images. We propose and analyse several natural colour image fusion metric configurations and formulations and define explicit colour preservation models to evaluate preservation of colour information within the multisensor fusion context. Results on realistic multisensor colour data demonstrate the robustness of several metric configurations and their ability to correctly rank fusion algorithms. We also show that colour information contributes modestly to perceived fusion success and can even be misleading if not considered within the wider context of overall image structure.
展开▼
