结合引导滤波和卷积稀疏表示的红外与可见光图像融合

为了解决红外与可见光图像融合时信息容易相互干扰、影响融合质量的问题,将引导滤波、高斯低通滤波与非下采样方向滤波器组相结合,提出一种新的图像融合方法。利用引导滤波和高斯低通滤波,将源图像分解为低频近似部分、强边缘部分和高频细节部分,并将高频细节部分进行非下采样方向滤波,进一步得到高频方向细节部分;对低频近似部分应用基于局部区域能量的融合规则,对强边缘部分提出一种基于卷积稀疏表示的融合规则,对高频方向细节部分提出改进的脉冲耦合神经网络的融合规则,得到相应的融合部分,并通过逆变换得到最终的融合图像。对多组红外与可见光图像的实验结果表明,算法得到的融合结果的主观视觉效果和客观评价指标均优于传统的图像融合方法,其客观评价指标中的标准差、信息熵、互信息、平均梯度和空间频率相比融合效果较好的基于离散小波变换和稀疏表示的融合方法平均提高20.28%、2.24%、47.41%、5.34%、8.02%。

Infrared and visible image fusion using guided filter and convolutional sparse representation

In order to solve the problem that the information from the source images is easy to interfere with each other which influences the quality of infrared and visible image fusion, a new image fusion method based on Guided filter, Gaussian filter and nonsubsampled directional filter bank was proposed. The low-frequency approximation components, strong edge components and high-frequency detail components were obtained by combining Guided and Gaussian filter. Then the high-frequency detail components were filtered to obtain the detail directional components with the use of nonsubsampled directional bank. The low-frequency approximation components were fused by a fusion rule based on regional energy and the strong edge components were fused by a strategy based on convolutional sparse representation. The detail directional components were fused by a rule based on improved pulse coupled neural network. Then the final fused results were obtained by using inverse transform through fusing the fused components. Experimental results show that the proposed algorithm outperforms traditional methods in terms of visual inspection and objective measures. Compared with the image fusion algorithm based on discrete wavelet transform and sparse representation, which possesses the better fusion effect in the traditional methods, the fusion quality indexes of the proposed method, such as Standard deviation(STD), Information entropy(IE), Mutual information(MI), Average gradient (AG) and Spatial frequency(SF) increased by 20.28%, 2.24%, 47.41%, 5.34%, 8.02% averagely.