CAJ | 학술논문

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。
해문침대홍외도상중함유비균균성조성화고사조성적퇴화모형，제출료일충기우각향이성(Perona Malik, PM)확산적신경망락비균균교정(PM-NN-NUC)산법。건립료관우비균균교정적겁소화모형。통과대신모형적최두하강방정화편미분방정적추도，가이간출PM-NN-NUC산법이용료신경망락교정화PM확산재려파과정중적상사성，불부직접용우산생신경망락교정적기망치，환작용우계산질대보장，이교정계수우반작용우PM적확산과정，경호지장 PM 확산화신경망락교정통일지결합재일기。통과대실제함조홍외도상진행실험，증명신모형가억제비균균조성，병방지도상산생퇴화。
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.