Imaging Though Simple Lens Based On Computational Photography

Modern imaging optical system generally by increasing the number of optical elements, introducing non-spherical surface even the freeform suface to eliminate system aberrations and improve image quality. This will lead to the optical system complexity increased, processing difficult and high cost, and it is also limited to alignment, technology, weight and other factors. In order to simplify the complexity of the optical system, a simple lens imaging technique based on computational photography is proposed, which is under the premise of ensuring the quality of the system. Combining optics with computer technology which is rapidly developed, through the image processing technology to restore blur images imaging with simple lens, reconstruct clear images and improve image quality. The conventional image restoration method is based on the assumption that the point spread function(PSF) of optical system is space-invariant, but the actual optical system because of the influence of aberrations, the PSF is changed with the field of view. In this paper, making an improvement on the existing sparse prior deconvolution algorithm, and proposing the variable parameter of the sparse prior non-blind deconvolution algorithm, and the image restoration. The main contents of the article as follows:First, the space variation characteristics of the optical system point spread function were analyzed, and put forward two acquisition methods of space-varying PSF. One is extracting in the process of optical system design, preprocessing by algorithm and retaining effective information. The second method is measuring the LSF of optical system through instrument, then caculate the PSF by algorithm according to the relations among the LSF, MTF and PSF. At last the way of interpolation are used to get spatially-varying PSF of the whole image.Second, blocking the blurred image overlapped, and each piece of image deconvolution with different PSF. Natural image gradients meet the sparse distribution, introducing sparse priors as a regularization term in deconvolution algorithm, setting the parameter which representing the image gradient sparsity as a variable value, and using iterative reweighted least squares(IRLS) to deconvolve a image. The gradient of recovered image in the iterative process is closer to a true gradient distribution, while avoiding excessive smoothing of image. The overlapped block restoration method can effectively reduce the ringing artifacts in image stitching. Image average method is used in image overlap region to make the image smoother when spliting the image.Finally, to verify the effectiveness of the algorithm, a digital simulation experiment is carried out on the simple lens system. The experimental platform is set up to measure the spatial variation PSF of a single lens, and the blurred image is reconstructed. The validity of the algorithm based on computational photography to improve image quality through simple lens is further verifiedExperimental results show that the proposed algorithm has good recovery effect on the simple lens optical system, and the image quality has been improved.