应用统计信号处理和模糊数学的图像融合算法

Image fusion algorithm based on statistical signal processing and fuzzy theory

针对多传感器图像在像素级上的融合问题,将模糊数学理论引入到图像融合模型。该模型假定理想的融合后的图像包含场景所有的信息;将它乘上一个模糊因子,再加上随机噪声,可用来描述某一个成像传感器中获得的场景图像;不同的传感器对应不同的模糊因子和噪声。在此基础上,提出了建立在非多尺度分解框架下的图像融合算法。它以各传感器获取的图像作为输入条件,应用统计信号处理中的EM算法,求出针对不同传感器的噪声参数和模糊因子,通过迭代估计出融合的图像。实验结果显示,该算法获得的融合图像的互信息和联合熵分别达到3.5079和24.732,均优于加权平均融合法、小波融合算法和Laplacian融合算法的融合质量。

For pixel-level multi-sensor image fusion, fuzzy theory is adopted to improve the fusion model. The proposed fusion model assumes that the ideal fusion image includes the entire information of scene. The field image from an imaging sensor can be described as two parts, one is the ideal image multiplied by a fuzzy factor, the other is stochastic noise. Each sensor has its corresponding fuzzy factor and parameters of noise level. Based on the frame of non-multi-scale decomposition, an image fusion algorithm is then proposed. Using the EM algorithm in statistics, the fuzzy factor and noise parameters of corresponding sensor can be derived. And then by iteration operation, the fusion image is finally estimated. Experimental results demonstrate that the mutual information and the union entropy of the fusion image by our algorithm reach 3.5079 and 24.732, respectively, which are superior to pixel average algorithm, wavelet fusion algorithm and Laplacian fusion algorithm.