利用脉冲耦合神经网络的图像融合

为了获得对同一场景更为准确、全面和可靠的图像描述,提出了一种基于脉冲耦合神经网络(PCNN)的图像融合方法。将多源传感器图像配准后的各个源图像用9/7小波变换的提升算法进行分解,从而得到各个源图像的低频分量和高频分量。对于低频分量,采用像素绝对值选大法进行融合;而高频分量则作为PCNN的输入,在迭代结束后,通过比较PCNN点火次数得到一系列融合子图像;然后,用9/7小波的提升算法将获取的一系列多尺度融合子图像进行反变换得到最终的融合图像。设计了可见光图像与红外图像的融合实验,对融合图像的熵、平均梯度、标准差、空间频率进行了定量比较。当使用标准源图像进行融合时,各值比使用传统小波变换与PCNN相结合的图像融合方法分别高0.0104,0.2459,0.1131和0.2846。

Image fusion based on pulse coupled neural network

In order to represent a scene exactly and entirely, an image fusion method based on Pulse Coupled Neural Network (PCNN) is proposed. After registering the images of multi-source sensors, obtained images are decomposed into several coefficients of low frequency and high frequency by using the 9/7 wavelet transform based on lifting scheme. The larger absolute gray values are selected to fuse low frequency images and the high frequency images are input to the PCNN, then a serial of fused sub-images can be obtained by comparing firing times after the iteration. Finally, the fused images are obtained by inversing transform using the 9/7 wavelet based on lifting scheme. By means of design of simulation experiments using visible and infrared images, the entropy, average gradient, standard deviation and space frequency are selected to evaluate the fused image. Obtained results show that the entropy, average gradient, standard deviation and space frequency of the fused image by using the novel fusion method base on PCNN are higher 0.010 4, 0.245 9, 0.113 1 and 0.284 6, respectively, than those by using the fusion method combining traditional wavelet and PCNN when a standard source image is used.