FABEMD和改进局部能量窗口的红外中波和长波图像融合

针对探测波段为3.7～4.8 μm的中波红外图像和探测波段为8～14 μm长波红外图像融合过程中存在场景对比度低，显著性目标不够凸出，伪影引入严重的问题，采用快速自适应二维经验模态分解（FABEMD)对红外中波和长波图像进行多尺度分解以得到二维内蕴模函数(BIMFs)和残余分量(Residual)。对于每一层内蕴模函数选用改进的局部能量窗口融合规则，首先配置好加权算子以增加区域窗口中心像素的能量占比；选用不同的加权算子，经实验验证能有效突出红外中波和长波图像的能量特征信息；其次充分利用内蕴模函数的相位信息，当相位相反时，采用能量加权平均的方式，以解决融合系数的正负符号极性难以确定的问题；当相位相同时，判断二者的能量差距并依据差距大小选择设定的融合规则，融合规则基于红外中波和长波图像的灰度差异特性设定。对于残余分量则利用红外中波图像和改进区域能量窗口的最大对称环绕显著性权重图指导基础层系数的融合，自适应的局部环绕窗口充分利用了低频显著性信息，对无用背景的抑制效果也相当出色，能够在复杂背景图像中突出显著性对象，最终得到细节信息丰富，对比度明显的指导图像。最后通过FABEMD的逆变化重构过程得到融合图像，对4组不同背景、不同大小的红外中长波图像进行主观和客观性能评价，4组图像均来自多波段红外采集系统且都经过严格配准并和7种相关算法进行对比实验，在主观性能上显著性对象突出、清晰度度高；客观性能上在平均梯度和空间频率这两个评价指标上性能优异，验证了该算法的有效性。

Infrared Mid-Wave and Long-Wave Image Fusion Based on FABEMD and Improved Local Energy Window

Aiming the scene contrast is low in the fusion process of the mid-wave infrared image with a detection band of 3.7 to 4.8 μm and the long-wave infrared image with a detection band of 8 to 14 μm, the saliency target is not enough to protrude, and the artifacts introduce serious problems. In this paper, Fast and Adaptive Bidimensional Empirical Mode Decomposition (FABEMD) is used to the multi-scale decomposition of infrared medium-wave and long-wave images to obtain two-dimensional intrinsic mode functions (BIMFs) and residual components (Residual). For each layer of bidimensional intrinsic mode function, this paper’s improved local energy window fusion rule is selected. First, a weighting operator is configured to increase the central pixel’s energy proportion for the regional window. In this paper, different weighting operators are selected, which have been verified to effectively highlight the energy characteristic information of medium-wave and long-wave images by experiments, and secondly, the phase information of BIMFs is fully used, when the phases are opposite, the energy weighted average method is used to solve the problem that the polarity sign of the fusion coefficient is difficult to determine. When the phases are the same, the energy gap is judged, and the set fusion rule is selected according to the size of the gap based on the grayscale difference characteristics of the infrared medium wave and long wave images. Using the infrared mid-wave image and improved the regional energy window of Saliency detection of maximum symmetric surround weight map guides the fusion of base layer coefficients for the residual components. The adaptive local surround window makes full use of the low-frequency saliency information and has a very good suppression effect on the useless background. It can highlight the saliency objects in the complex background image, and finally obtain the guidance image with rich details and obvious contrast. Finally, the fusion image is obtained through the inverse reconstruction process of FABEMD, and subjective and objective performance evaluations are performed on five sets of infrared medium and long-wave images with different backgrounds and different sizes. The four sets of images are all taken from multi-band infrared acquisition systems and are strictly registration and comparative experiments with 7 related algorithms. In terms of subjective performance, salient objects is standing out and the clarity is high, the objective performance is excellent in two evaluation indicators of average gradient and spatial frequency, the effectiveness of this algorithm is verified.