Research On The Method For Different-source Image Fusion Quality Evaluation

As the feedback link of the fusion algorithm,the image fusion quality evaluation has the key role of evaluating the fusion algorithm and guiding the development of the fusion algorithm.At present,the research on the quality evaluation of multi-source image fusion has been improved,and many objective evaluation metrics of multi-source image fusion with excellent performance have been proposed.On this basis,by referring to the quality evaluation method of multi-source image fusion,and combining the features of different structures and the complex noise types of different-source images,representative infrared image(IR),synthetic aperture radar image(SAR)and visible image as the research object,two kinds of quality evaluation metrics of different-source image fusion on noise level and structure preservation are put forward.The multi-source image fusion quality evaluation metric is introduced.This paper classifies the existing multi-source image fusion quality evaluation metric,expounds the principles and calculation steps of several common fusion quality evaluation metrics,and discusses the limitations and existing problems of the above fusion quality evaluation metrics.The research on objective evaluation metric of multi-source image can provide reference for the construction of different-source image fusion objective evaluation metric.An evaluation metric of different-source images fusion based on blind noise level estimation is studied.Firstly,selecting a weak texture image block of the visible image according to the complexity of the scene,wherein the selection mode is based on the gradient and statistical characteristics of visible image blocks;Then the weak texture image blocks of visible are corresponding to the fusion image pixel by pixel as the weak texture blocks of the fusion image;Finally,blind noise level estimation based on principal component analysis(PCA)is used on the weak texture blocks of the fused image to obtain the noise level of the fused image.The proposed blind noise estimation metric and the existing blind noise estimation metric are used to evaluate the noise of four classic fusion algorithms.Experimental results show that the proposed blind noise estimation metric has higher subjective consistency in the noise estimation of SAR and visible image fusion,and its accuracy and robustness are better than the existing blind noise estimation metrics.An evaluation metric of different-source image fusion structure preservation based on phase congruency mutual information is studied.First,the structural feature maps of the source image and the fused image are extracted by phase congruency.Because the phase congruency is easily affected by noise,based on the idea of non-local mean(NL-Means),the energy in each direction is denoise according to the maximum likelihood estimation weight of the original image block,and the denoising phase congruency is obtained;Then,phase congruency maps of all the source images are large enough to obtain a comprehensive denoising phase congruency map;Finally,calculating mutual information of the source image comprehensive denoising phase congruency map and the fused image denoising phase congruency map,and the different-source image fusion structure preservation metric is obtained.The proposed structure preservation metric and existing structure preservation metric are applied to the structure preserving analysis of four classical fusion algorithms.The experimental results show that the proposed structure preservation metric has higher subjective consistency in the evaluation of structural features of different-source image fusion,and its accuracy and robustness are better than the existing structure preservation metric.