基于PM扩散的红外焦平面阵列神经网络非均匀校正算法  

Neural Network Non-uniformity Correction for Infrared Focal Plane Array Based on Perona Malik Diffusion

作　　者：杨硕[1] 赵保军[1] 毛二可[1] 唐林波[1] 

机构地区：[1]北京理工大学电子与信息学院,北京100081

出　　处：《电子与信息学报》2013年第11期2744-2750,共7页Journal of Electronics & Information Technology

摘　　要：该文针对红外图像中含有非均匀性噪声和高斯噪声的退化模型,提出了一种基于各向异性(Perona Malik,PM)扩散的神经网络非均匀校正(PM-NN-NUC)算法。建立了关于非均匀校正的极小化模型。通过对新模型的最陡下降方程和偏微分方程的推导,可以看出PM-NN-NUC算法利用了神经网络校正和PM扩散在滤波过程中的相似性,不仅直接用于产生神经网络校正的期望值,还作用于计算迭代步长,而校正系数又反作用于PM的扩散过程,更好地将PM扩散和神经网络校正统一地结合在一起。通过对实际含噪红外图像进行实验,证明新模型可抑制非均匀噪声,并防止图像产生退化。A new Neural Network Non-Uniformity Correction （PM-NN-NUC） algorithm is proposed for InfraRed Focal Plane Array （IRFPA） based on Perona Malik （PM） diffusion for the situation of degradation model both containing fix pattern noise and Gaussian noise in infrared image. A minimize model is established concerning Non-Uniformity Correction （NUC）. It can be seen that PM-NN-NUC uses a similarity in the filtering process on Neural Network Non-Uniformity Correction and PM diffusion, and not only generates the expectation directly but also calculates the iterative step. Correction coefficient reacts on PM diffusion process and combines with PM diffusion and Neural Network Non-Uniformity Correction uniformly. The results of real infrared thermal image show that the proposed algorithm eliminates the fixed pattern noise effectively, but also has excellent performance for the image degraded with fade-out.

关 键 词：图像处理 神经网络非均匀校正法 各向异性扩散 最陡下降方程 偏微分方程 图像退化 

分 类 号：TN911.73[电子电信—通信与信息系统]