潜在低秩表示下VSM联合PCNN的红外与可见光图像融合

Infrared and visible image fusion of VSM and PCNN under potential low-rank representation

为了更好的保存红外与可见光图像中的细节轮廓和对比度等信息,提出了一种潜在低秩表示(LatLRR)下视觉显著映射(VSM)和脉冲耦合神经网络(PCNN)相结合的图像融合方法。通过LatLRR将红外图像和可见光图像分解为低秩部分和显著部分,并分别使用不同的融合策略对得到的低频和高频层进行融合。低秩部分上,先计算图像低秩部分的视觉显著强度,并使用基于VSM的融合策略来融合图像低秩部分,以保留更多轮廓信息;显著部分上,使用梯度域PCNN融合策略来融合显著部分。舍弃二者的稀疏噪声,通过对融合后的低秩部分和显著性部分进行叠加,得到最后的融合图像。与其他经典的融合方法进行对比验证,融合图像的相关差异和(SCD)、结构相似性(SSIM)、融合质量(Q^(AB/F))等多种客观指标均有所提升,图像细节信息丰富,清晰度高,具有良好的可视性。

In order to better preserve the detail contour and contrast information in infrared and visible images, an image fusion method based on the combination of visual significance mapping(VSM) and pulse coupled neural network(PCNN) under latent low-rank representation(LatLRR) was proposed. The infrared image and visible image were decomposed into low-rank part and significant part by LatLRR, and different fusion strategies were used to fuse the obtained low-frequency and high-frequency layers, respectively. For the low-rank part, the visual significant intensity of the low-rank part of the image was calculated, and the fusion strategy based on VSM was used to fuse the low-rank part of the image to retain more contour information. For the significant part, the gradient domain PCNN fusion strategy is used to fuse the significant part. The sparse noise of the two images is discarded and the low-rank part and the significance part are superimposed to obtain the final fusion image. Compared with other classical fusion methods, the relevant differences and(SCD), structural similarity(SSIM), fusion quality(Q^(AB/F)) and other objective indicators of the fusion image are improved, and the image has rich details, high definition and good visibility.