Research And Implementation Of Super-resolution Algorithm Based On Light Field Image

With the development of imaging technology,the traditional imaging method can only record the projection of the scene on the plane,which limits the further development of optical imaging.With the deepening of the theoretical research of light field imaging and the gradual improvement of the hardware technology level,light field imaging subverts the traditional imaging mode,can completely record the light field information of the scene,and can realize the functions of refocusing and depth estimation through subsequent processing.Therefore,it has great development potential.As one of the main methods of light field acquisition,the light field camera based on microlens array is limited by its own structure and faces the problem of low spatial resolution.The thesis takes the light field image as the research content,realizes the extraction of the light field sub aperture image,and carries out the research on the spatial super-resolution reconstruction of the light field image.Firstly,aiming at the problem that the image taken by the light field camera can not be used directly,the thesis realizes the extraction of the light field sub aperture image.Use the light field camera to collect the uniform white scene,calibrate the center of the microlens with the help of the obtained white image,and obtain the parameters such as the center coordinates of the microlens.Considering the limitations of technology,the microlens array may not be strictly aligned with the image detector.The alignment can be realized by rotating,scaling and translating the captured scene image using affine transformation.Because the macro pixels are arranged in a hexagon,it is also necessary to carry out orthogonal transformation on the macro pixels,and then extract and reorganize the sub pixels in the macro pixels to complete the extraction of the light field sub aperture image,so as to lay the foundation for the subsequent super-resolution reconstruction of the light field image.Then,aiming at the problem of low spatial resolution of light field image,a spatial super-resolution reconstruction algorithm based on light field image is realized.By comparing the structural characteristics of light field space with the timing characteristics of video sequence,a simple light field sequence generation method is proposed.With the help of more mature video super-resolution reconstruction idea,a light field superresolution reconstruction method based on generating countermeasure network is designed.Finally,the super-resolution reconstruction algorithm is verified on different data sets.The experimental results show that this method can reconstruct high-resolution images with good subjective evaluation indexes,compared with ESRGAN,PSNR is increased by 1.14 d B and SSIM is increased by 0.023.