应用小波变换的自适应脉冲耦合神经网络在图像融合中的应用

Application of adaptive PCNN based on wavelet transform to image fusion

设计并实现了一种适用于红外与可见光图像融合的基于小波变换的自适应脉冲耦合神经网络(PCNN)融合技术。首先,对融合的两幅图像进行小波分解得到两组多尺度图像。然后,在小波域充分利用PCNN的同步激发特性,进行PCNN的融合策略设计;使用不同频率下小波系数的局域熵作为PCNN对应神经元的链接强度,经过PCNN点火获得参与融合图像在小波域中的点火映射图;根据点火时间计算点火映射梯度图,再通过判决选择算子,选择点火时间梯度最大的小波系数作为融合系数。最后,对融合后的小波系数进行重构生成融合图像。进行了两组图像融合实验,结果显示,在迭代次数为50次时,与经典小波方法相比,两组实验结果的熵分别提高1.1%,0.7%;平均梯度分别提高8.3%,3.7%;空间频率分别提高2.5%,1.5%;标准差分别提高1.9%,0.6%;交叉熵分别缩小5.6%,4.9%,结果表明本文方法用于红外与可见光图像的融合十分有效。

A fusion method of infrared and visible light images based on Pulse Coupled Neural Network （PCNN） and wavelet transform is studied. Firstly, the two original images are decomposed by wavelet transform, then, a fusion rule in the wavelet domain is given based on the PCNN. This algorithm uses the local entropy of wavelet coefficient in each frequency domain as the linking strength, then its value can be chosen adaptively. After processing PCNN with the adaptive linking strength, new fire mapping images are obtained. According to the fire mapping images, the firing time gradient maps are calculated and the fusion coefficients are decided by the compare-selection operator with firing time gradient maps. Finally, the fusion images are reconstructed by wavelet inverse transform. Two groups of experiments are undertaken for the fusion of visible and infrared images, results indicate that when the numbers of iterations are 50 times, the entropy has increased by 1.1% and 0.7%; the average grads by 8. 3% and 3. 7%; the spatial frequencies by 2.5% and 1.5%; the standard deviation by 1.9% and 0.6%, respeetively; and the cross-entropy has reduced by 5.6% and 4.9%, respectively as comparing with that of classical wavelet method. These results show that proposed method has improved the details of fused images and is suitable for fusing visible and infrared images.