Research On Light Field Super-resolution Reconstruction Algorithm Based On Spatial Structure Features

With the rapid development of mobile interactive devices and the gradual popularization of 5G networks,people are increasingly eager to obtain a higher-quality information interaction experience.From texts,images,and voices to the increasingly popular short videos,the carriers of information are gradually changing from simple to complex,from one dimensional to higher dimensions.Three dimensions are the dimensions of the objective and real world,and three-dimensional information can more intuitively reflect the information of this world.Light field imaging is a technique for obtaining high-dimensional visual information from the real world.Compared with traditional two-dimensional imaging,it can expand the acquisition of light information by acquiring the spatial and angular distribution of light,and calculate the required image through data processing methods such as transformation and integration.Light field imaging has been successfully applied in refocusing after the shooting,semantic segmentation and depth estimation based on one shot.Today's mainstream micro-lensarray-based light field cameras provide a cost-effective way to obtain light fields.However,due to its own structural limitation,the acquired light field image has a tradeoff between spatial and angular resolution,and faces the problem of low image resolution.In order to solve this problem,the research on spatial super-resolution reconstruction of light field images is carried out.The main work of this thesis is as follows:(1)After analyzing the spatial structure features of the light field and the geometric relationship between the images from different viewpoints,and comparing the similarities and differences with the temporal structure features of video sequences,a simple and efficient method for generating light field image sequences is designed.(2)Based on the analysis of the light field spatial structure features,combined with the above-mentioned light field image sequence generation method,the global and local spatial structure feature fusion and optimization are performed for the light field image sequence,and a global and local spatial structure feature-based,multi-branch,and endto-end light field image super-resolution reconstruction method is proposed.Experimental results show that this method can effectively extract and use the spatial structure features of light field images to reconstruct high-resolution images with good results.Compared with several mainstream light field super-resolution reconstruction algorithms,the peak signal-to-noise ratio has a 0.8 dB to 1.26 dB improvement.(3)To further utilize the spatial structure features of the light field,by analyzing the features of the epipolar plane image(EPI),using the special texture features of the EPI,and using the residual block constructed by 3D convolution,an EPI-based light field super-resolution reconstruction optimization method is proposed.The experimental results show that the method can make fuller use of the spatial structure features of the light field,and the reconstruction result is good.Based on the reconstruction method proposed in(2),a gain of 0.06 dB to 0.4dB is further obtained.