Feature transfer method for infrared and visible image fusion via fuzzy lifting scheme

Abstract

To reserve or enhance the integrity of regional heterogeneous features, a feature transfer image fusion method based on Fuzzy lifting (FTF) is proposed, which includes a fuzzy lifting stage, a feature transfter process and a new reference composite feature evaluation index. Different from traditional fusion methods which usually adopts an undifferentiated representation and a global unified fusion method for the original image, as for the infrared and visible image fusion, imaging characteristics and visual interest features at different physical locations are different, so the important local features on each images may be lost or weaken during the fusion process. The above problems can be solved by proposed FTF method from two aspects. Firstly, in the fuzzy lifting stage, the Fourth order Partial Differential Equation (FPDE) and fuzzy region rules are crossly performed to remove redundant information and highlight the main features of the region. Secondly, a feature transfer model is established to perform a more efficient feature transfer on the lifted feature map. Besides, considering the shortcomings of existing indexes in feature description, a new ComPosite Feature metric (CPF) for fused images are also proposed. Qualitative and quantitative comparisons are made between the FTF method and other eight state-of-art methods on TNO image fusion dataset. The experimental results show that the salient thermal radiation targets and clear visible details can be represented by the final fused images, which is conducive to the monitoring and analysis. In summary, the proposed FTF method is more effective, and has better ability to catch most of the salient feature from source images than that of state-of-art methods.