A Novel Heterogeneous Fusion Algorithm Using Total Variation Model for Neutron and X-ray Radiography

The methodology of image fusion combines correlative information from several images into a single image which facilitates better visual perception and higher accuracy for further processing task than each of the target images. This fusion process is computationally expensive for large scale images. A parallel fusion scheme leads to Neutron radiography (NR) as well as X-ray radiography (XR) incorporating the Newton bucketing method along with depreciation effective on total variation scheme of anisotropic diffusion of 2nd order devised on Multi-core CPU-GPU based CUDA compatible platform, presented here. This scheme is unique and stable in terms of Euler-Lagrange solutions of the second-order anisotropic diffusion. Discretization method based on finite difference is used to establish precise and efficient numerical approximation whereas dependency of parameters also demonstrated in our proposed model. This scheme generates a sharp and enhanced fused image that combined smooth information of the NR and details of XR image simultaneously without noise and blur artifacts of the source images. In order to perform an efficient parallel fusion, all kernel functions have been implemented on GPU in this proposed scheme. This CPU-GPU based model makes a major difference on the basis of performance acceleration related to other sequential schemes devised on CPU. Experimental results exhibit performance acceleration and provide superior fusion outcomes than available approaches with regard to both quantitative and optical perceptivity.