4D Light Field Space Information Super Resolution Algorithm Research

Because it can obtain the spatial information and angle information of scene synchronously,the light field camera has important application value in depth estimation,target recognition,3D video acquisition and other fields.The angle information of the light field camera is acquired at the expense of spatial resolution.In order to make the light field image not limited by spatial resolution in the later application,we study the super resolution reconstruction algorithm of light field image spatial information based on deep learning.The specific research work and innovations are as follows:1.A light field super-resolution reconstruction network based on angle difference enhancement is built in this paper.In the shallow layer feature extraction,the Multi-branch residual block is used to extract the inherent structure information in the 4D light field;In the deep feature extraction part,enhanced angular deformable alignment module is designed to enhance the feature mining of each side view,so that the angle difference between views can be better learned;Angle information is well fused and encoded into each view feature.In the process of data reconstruction,the simplified residual feature distillation module is used to make the shallow and deep information in feature fusion more directly.The performance of the proposed network is verified on five public light field datasets,and the proposed algorithm obtained high-resolution light field subaperture images with higher PSNR and SSIM values.2.A multi-layer sensing super-resolution reconstruction network is built.In the proposed network,the alternate Residual atrous spatial pyramid pooling and Multi-branch residual blocks are used to extract shallow features,then Multi-level perception residual module and Global feature updating block and Local feature updating block can be used to extract deep features.This module takes advantage of the correlation between all views,while maintaining the parallax structure of the light field view,and finally the enhanced residual feature distillation module and pixel shuffle module are used to complete the data reconstruction.On the premise of ensuring the feature fusion effect,the calculation cost is effectively saved.The experiments are carried out from the aspects of validity verification of each network module,comparison of subjective visual effects,comparison of quantitative evaluation results and algorithm complexity.The performance of the proposed network is verified on five public light field data sets.The proposed algorithm obtains high-resolution light field sub-aperture images with higher PSNR and SSIM.