Research of Multi-scale Decomposition-based Noisy Multispectral Image Fusion Methods Efficiency
Abstract:
The article considers algorithms of multiscale decomposition under the influence of additive noise in one of the channels of a multispectral vision system. The difference between pyramid-based and wavelet-based decomposition-reconstruction methods is shown. Structural schemes for the realization of different image fusion strategies are presented, and their advantages and disadvantages are described. To estimate the fused image quality the authors applied the complex integral-multiplicative index of digital grayscale image quality that operates with such partial indices as signal-to-noise ratio, local contrast, and high-frequency signal component root-meansquare deviation. The dependences of numerical values of the quality index on the standard deviation of additive white Gaussian noise in the visible range channel of the multispectral vision system are presented. Thus, when ?RMS > 10, for the most considered image fusion algorithms, a decrease of the integral-multiplicative quality index by 2…10 times is observed, which under a priori unknown observation conditions confirms the inexpediency of enhanced vision systems working with a constant image fusion from all channels. The known high-frequency multiscale decomposition results fusion methods disadvantages are described. The approach to combat these disadvantages, which consists of fusion only of those images whose quality index is not lower than the threshold one, is proposed.
