Research On Infrared Image Super-resolution Reconstruction Based On Compressive Sensing

Compared with visible-light imaging systems,the reliability and stability of infrared imaging systems are still far from satisfactory.Its performance is greatly affected by the environment,which can not be completely solved by hardware.Therefore,through the software method,namely super-resolution reconstruction,from the observed low-resolution image sequence of one or more frames,high-quality or high-resolution im-age can be obtained,and image blur and noise can be removed at the same time,so it is more practical to carry out post-processing of infrared images.At present,infrared image super-resolution processing usually uses the spatial method represented by gray-scale transformation and histogram equalization,and transform domain method based on wavelet transform and Fourier transform.As a hot spot in recent years,compressive sens-ing theory provides a new idea for the research in many fields with the idea of signal sparse representation,compression coding,and high-quality reconstruction.The image super-resolution problem is regarded as the process of reconstructing the original signal from the sampled signal,which opens up a broad research space for image super-resolution reconstruction.Starting from the theory of compressive sensing,this paper mainly studies the method of high-quality reconstruction of infrared images from the perspective of removing in-frared image blur and improving image resolution.The specific research content includes the following aspects:?1?Research on the basic theories of compressive sensing,including sparse repre-sentation,observation matrix,and reconstruction algorithm.It focuses on the minimum l₁norm method,greedy algorithm and iterative threshold method,and analyzes and com-pares various algorithms from three aspects:sparse solution,iterative process,and algo-rithm complexity,which provides a theoretical basis for subsequent research.?2?Analyzing of the types of infrared image noise,this paper studies the infrared image reconstruction under salt and pepper noise and various blur conditions,and proposes an infrared image deblurring method using overlapping group sparsity and l_pquasinorm.This method uses the l_pquasinorm to replace the l₁norm in the traditional model,so that the image can be reconstructed with better sparseness characteristics.The reconstruction effect under different salt and pepper noise,Gaussian blur and mean blur conditions has been significantly improved.?3?A method combining Shearlet transform and total variation to remove Gaussian noise in infrared images is proposed.For image denoising,when the Shearlet transform is used alone because it includes a down-sampling process and does not have translation invariance,the Gibbs phenomenon may occur;similarly,when the total variation is used alone because in each pixel both the horizontal and vertical gradients are calculated,so there will be a staircase effect in non-edge areas.Therefore,the combination of the two can give full play to the characteristics of Shearlet and total variation in image geomet-ric features and edge protection,reduce Gibbs phenomenon and staircase effect,improve image denoising effect,and achieve the high-quality reconstruction of infrared images.?4?A super-resolution reconstruction method based on four-direction fractional total variation and l_pquasinorm is proposed.Integer-order total variation can maintain the dis-continuity and the structure of the image,but the block effect problem often occurs when it is used for super-resolution reconstruction.Fractional total variation can well deal with non-local features of the image,such as edges and textures.Using it for super-resolution reconstruction can alleviate the block effect problem.Also,to make full use of the gradient information of the pixels in multiple directions to improve the performance of denoising and blurring,the four-direction gradient is applied to the fractional total variation;the l_pquasinorm can also bring better sparse characteristics,further improve the quality of infrared image super-resolution reconstruction.?5?A super-resolution reconstruction method based on four-direction and high-order overlapping group sparsity total variation is proposed.Based on the previous research,four-direction total variation is applied to the overlapping group sparsity total variation,and the correlation between pixels is more fully used in the process of image denoising,deblurring,and super-resolution reconstruction.The high-order gradient,in addition to using the gradient information in the horizontal and vertical directions,also considers the relationship between the gradient information between three adjacent pixels in the hori-zontal and vertical directions and the gradient information between adjacent four pixels.It can suppress the staircase effect in the super-resolution reconstruction process and im-prove the quality of infrared image super-resolution reconstruction.