Noise reduction for gradient-based multispectral image fusion

We propose a new noise reduction algorithm for gradient-based fusion of multispectral images degraded by noise. The gradients resulting mainly from noise are less included in the fusion process than those from salient features according to a minimum mean square error (MMSE) criterion. The selection of an appropriate stationary support region for the criterion is studied. Experimental results show that the proposed method is superior to pre- or post-processing methods of noise reduction based on MMSE criteria. The fused result of multispectral iris images is presented for an application to iris recognition.