Research On Super-resolution Reconstruction Algorithm Of Infrared Image

Infrared image is very common in life and plays an important role in many fields,but compared with color image,infrared image has poor stereoscopic sense and insufficient saturation,and the current infrared imaging technology has limited detection ability,and can not have high spatial resolution like visible light imaging,so for direct observation of human eyes,infrared image is far from meeting the visual needs 。 In this paper,aiming at the shortcomings of infrared image,we improve the regularization algorithm,and put forward the super-resolution reconstruction algorithm of double regular sub-pixel convolution infrared image and the super-resolution reconstruction algorithm based on depth plug-and-play.The main research contents include:First of all,because the details of the image reconstructed by the traditional regularization algorithm are fuzzy and the edge is not clear,this paper improves the traditional regularization super-resolution reconstruction algorithm: first,the image is enhanced with multi-scale details before reconstruction;then,another regularization term and adaptive parameter are added to the reconstruction target function to enhance the image edge.According to the experimental data,the subjective and objective evaluation indexes of the image obtained by the improved regularized infrared image reconstruction algorithm are improved.Secondly,due to the lack of prior information,insufficient feature extraction and the complexity of using the interpolated image as the reconstruction input,a dual regularized sub-pixel convoluted infrared image(MPSR)super-resolution reconstruction algorithm is proposed to solve the above problems,which combines regularization with sub-pixel convolution Firstly,regularize the input image to enhance the details and edges;secondly,extract and activate the features;finally,rearrange the pixels with sub-pixel convolution layer to get the high-resolution image,which not only improves the image quality but also improves the algorithm training speedFinally,the single image super-resolution(SISR)method is mainly designed for the widely used Bicubic degradation.For the super-resolution low-resolution(LR)image with arbitrary fuzzy kernel,there are still fundamental challenges.In this paper,we use the idea of migration learning to migrate the newly proposed depth plug-and-play model to thesuper-resolution reconstruction algorithm.According to the experiment,we get It can be seen from the data that the improvement proposed in this chapter is effective.Compared with other algorithms,the PSNR and SSIM values are significantly improved,and the performance of any fuzzy kernel is stable,the improvement of image detail information and edge preservation are significantly better than other algorithms.Moreover,it can effectively reduce noise,improve the overall brightness of the image,and the subjective visual effect is better than before.