Research On Super-resolution Reconstruction Method Of Microscanning Infrared Image

With the wide application of infrared images in military defense,resident life and other fields,improving the resolution of infrared images has become the focus of research.Micro-scan infrared imaging technology is very helpful to improve the resolution of infrared images,but due to the limitations of hardware conditions,it is difficult to continue to improve the image resolution of hardware.Super resolution reconstruction technology can process multi-frame low resolution images with complementary information.And without changing the conditions of the hardware system,one or more clear high-resolution images are reconstructed.Therefore,using existing multi-frame low-resolution infrared image information for super-resolution reconstruction and improving the quality of image reconstruction has become a hot topic of research.This article takes micro-scanning infrared images as the main research goal,starting from the perspective of reconstruction and learning methods to conduct infrared image super-resolution research.First of all,the theoretical basis of micro-scanning infrared super-resolution imaging and existing technologies are theoretically studied,including infrared imaging technology,micro-scanning technology,image super-resolution reconstruction method and evaluation criteria.Subsequently,according to the characteristics of micro-scan infrared images,the multi-frame image super-resolution algorithm-convex set projection algorithm(Projection Onto Convex Sets,POCS)was improved,including motion estimation and image edge enhancement.Among them,in view of the characteristics of sub-pixel displacement and rotation of micro-scanned infrared images,the phase method is selected as the study,and the window function and Fourier-Mellin transform are used to reduce the effect of aliasing and improve the accuracy of motion estimation.In post-processing,a learning-based method is introduced to reduce the blurring effect caused by the point spread function and improve the ability to maintain the edge and detail of the infrared image.Finally,based on the learning-based single-frame reconstruction method and the video super-resolution reconstruction method,using the combination of single-frame and multi-frame super-resolution ideas,a super-resolution reconstruction for 4-frame low-resolution image input was designed using a recursive back-projection network.Infrared image sample library,and increase the details through the calculated residuals to obtain high-resolution images with clear edges and rich details.Multiple experimental results show that in the motion estimation part,the accuracy of the motion estimation algorithm improved and applied in this paper can effectively improve the reconstruction effect of the POCS algorithm.Compared with the traditional reconstruction algorithm,the introduction of post-processing based on learning can effectively improve the edge blur phenomenon after the reconstruction of the POCS algorithm and improve the sharpness of the edge contour.Compared with other methods,super-resolution experimental results of multi-frame infrared images based on learning have clear and obvious advantages in subjective and objective vision.The reconstructed image subjectively has clearer edge details,and the image effect is close to the original real image.It also has a certain improvement in the objective evaluation standard peak signal-to-noise ratio and structural similarity.