Research On Super-resolution Reconstruction Algorithm Of Light Field Image Based On Deep Learning

Light field cameras can simultaneously record the position information and angle information of the spatial light,and have significant advantages in digital refocusing,full-focus image acquisition,and depth estimation.However,the recording of multi-dimensional information is based on the premise of sacrificing the spatial resolution of the light field image.In order to make full use of the many advantages of the light field camera and overcome the problem of limited image resolution,it is of great significance to study the light field super-resolution reconstruction technology.Combining the advantages of deep learning in the field of image processing,this thesis carries out research on the super-resolution reconstruction algorithm of light field images based on deep learning.The main work and innovations are as follows:1.A GAN network light field image super-resolution reconstruction algorithm based on block artifact for data preprocessing is proposed.GAN network is often used for super-resolution reconstruction of two-dimensional images.When it is applied to super-resolution reconstruction of four-dimensional light field,four-dimensional data needs to be preprocessed.Based on the idea of light field coordinate transformation,this paper proposes a four-dimensional light field preprocessing algorithm based on block effect evaluation.By comparing light field macro pixel block effect values to find clear areas,and then realize the transformation of four-dimensional light field information into two-dimensional images,and finally transform The latter two-dimensional image is used as the input of the GAN network to achieve super-resolution reconstruction of the light field.The experimental results show that the data preprocessing algorithm proposed in this paper can obtain higher-quality high-resolution images than the direct projection and dimensionality reduction of the four-dimensional light field.The quantitative evaluation results also demonstrate the effectiveness of the algorithm.2.Propose a super-resolution reconstruction algorithm of light field image based on deep recursive residual network.In order to improve the performance of the residual light field super-resolution reconstruction network,this paper improves the network and proposes a light field image super-resolution reconstruction algorithm based on a deep recursive residual network.The algorithm uses recursive area operations to deepen the network structure.Weight sharing is used in the recursive module to reduce model parameters and at the same time make full use of the deep-level information of the image,and ultimately improve the quality of the image after super-resolution reconstruction.Compared with the residual light field super-resolution reconstruction network,the high-resolution image obtained by the improved network has a clearer visual texture.Quantitative evaluation shows that the peak signal-to-noise ratio and structural similarity of the super-resolution reconstructed image obtained by the proposed algorithm are increased by 0.26 db and 0.3 respectively.