A Dimensionality Reduction Method for the Fusion of NIR and Visible Image

Abstract

Fusing Near-infrared (NIR) and visible images aim to provide more detailed images for human inspection or other computer vision applications. Image fusion relies heavily on two components: measuring activity levels and assigning weights. The paper employs principal component analysis network (PCANET) to fuse Near-infrared (NIR) and visible data using an image pyramid. First, utilize a PCANET, an ultra-compact neural network for deep learning, to quantify action and give relative importance to infrared and visible images. PCANET's functional level assessment is superior in expressing key concepts such as NIR object processing and detailed visual representation. Second, the image pyramid divides the weights and source images into several scales and then applies the weighted-average fusing algorithm to the combined data. In the end, reconstruction produces the fused image. More than eighty test image pairings across two datasets confirmed the efficacy of the proposed approach. The experimental results reveal that the proposed strategy beats six benchmark approaches for subjective user evaluations and objective assessment metrics.