Research On Infrared Image Blind-pixel Compensation And Super-resolution Reconstruction Algorithm Based On Generative Adversarial Networks

Infrared images have great application value and development space in the fields of military detection,civilian surveillance and medical diagnosis.However,due to the limitation of materials and production processes of the current infrared imaging system and the influence of the external environment,infrared images taken will be affected by non-uniformity,and infrared blind-pixel will be generated.The image resolution obtained by the imaging system is low.How to realize the blind-pixel compensation and super-resolution reconstruction of infrared images has always been a key research issue in infrared image processing.Therefore,the paper conducts in-depth research on infrared blind-pixel compensation and super-resolution reconstruction,and proposes blind-pixel compensation and super-resolution reconstruction algorithms for infrared images based on generative adversarial networks.The main contents of this article are as follows:(1)Aiming at the problem of missing or over-detection of the traditional blind-pixel detection algorithm,an adaptive threshold detection method is introduced.The 3σ windowing method is used to implement blind-pixel detection of infrared blind-pixel images.To some extent,the problem of inaccurate detection of blind-pixel has been alleviated.(2)Aiming at the problem that traditional blind-pixel compensation algorithms have poor adaptability to clustered blind pixels,a generative adversarial network is introduced into the infrared image blind-pixel compensation.The generator adopts Encoder-encoder structure,the discriminator adopts Markov chain structure,and the gradient penalty term is introduced into the loss function to improve the stability of training.At the same time,the blind-pixel compensation effect is improved by Poisson image blending algorithm and iterative method.The algorithm makes full use of the image feature extraction capability of the convolutional neural network and the accurate prediction of the pixel gray value of the generative adversarial network,breaking through the limitations of the poor adaptability of traditional algorithm to clustered blind pixels.(3)Aiming at the problem of poor utilization of the spatial-temporal correlation of infrared image sequences by traditional algorithms,the optical flow motion estimation method and adaptive motion compensation method are used to achieve motion compensation between infrared image sequences.It laid the foundation for the subsequent infrared image super-resolution reconstruction task.(4)Aiming at the problem of poor reconstruction effect of traditional infrared image super-resolution reconstruction algorithms when performing high-magnification tasks,a generative adversarial network was introduced to perform super-resolution reconstruction of infrared images.The structure contains a deep residual network,which solves the problem of network degradation caused by too deep network structure,and achieves a good high-magnification reconstruction task.