Research On Super-resolution Reconstruction Based On Light Field Camera Image

With the continuous development of imaging equipment,light field imaging has become a technology that can capture richer visual information from real scenes.Compared with traditional ones,light field cameras can collect ray of light information in space from all directions,and can capture the four-dimensional information in the current scene with single exposure.This special feature makes light field cameras fit for variety of applications: re-focusing after image capture,depth estimation based on light field information,and three-dimension rendering of light field camera images.However,due to the special design structure of the light field cameras,its imaging resolution is limited by the sensor resolution.There is a trade-off between the angular resolution and spatial resolution.Light field cameras exhibit a very limited spatial resolution,which is insufficient to meet the application requirements in the current market.Aiming at the problem of low spatial resolution of light field camera images,this paper conducts research on super-resolution reconstruction algorithms based on light field camera images.Aiming at the calculation method of the self-similarity coefficient of the light field sub-aperture image,the relationship between the self-similarity coefficient and the image compression ratio obtained by fractal image coding is compared through design experiments.Based on this experiment,the correctness of the calculation method of the self-similarity coefficient with the light field sub-aperture image is determined,and the hypothesis that the self-similarity does exist in the light field sub-aperture image is verified.On this basis,a non-local mean constraint is proposed,and redundant information generated by the self-similarity of the light field image itself is used to supplement the image details.Based on the variational super-resolution framework,a super-resolution reconstruction algorithm based on non-local mean constraints is proposed.Then an experiment is designed to compare this one with other reconstruction algorithms.The experimental results show that the proposed algorithm can effectively recover the detailed texture information of light field images and obtain better reconstruction results.Because different sub-aperture images have unequal perspectives,there is a slight occlusion at the edges of the objects,which will affect the result of block matching among sub-aperture images.This paper conducts a set of experiments,using the real disparity map of the light field instead of the block matching result.Compared with the original one,it is found that the real disparity map significantly improves the accuracy of the reconstruction result.It also proves that the effect of image block matching results affects the reconstruction consequent.In view of the conclusions above,a regularization term based on adaptive matching is proposed due to the graph-based super-resolution reconstruction algorithm.One index of confidence is added to the matching image blocks.The more accurate the matching of the image blocks,the greater the role played in the constraints.While reducing the effect of mismatch on the stability of the algorithm,the reconstruction quality of the algorithm is improved.Finally,the subjective observation and objective indicators are used to compare and analyze the algorithm proposed in this paper with the existing ones.The experimental results show that the algorithm proposed in this paper improves the quality of the reconstructed image,and the restoration of the image structure as well.The peak signal-to-noise ratio is increased by an average of 0.25 dB,and the structural similarity image measure is increased by about 0.0017.