Research On Integral Imaging Reconstruction Based On Deep Learning

The advantage of integral imaging technology is that the observer does not need to wear visual aids during viewing,has good effect during stereoscopic image display,and will not produce visual fatigue during viewing.At the beginning of the advent of this technology,it has attracted the attention of many researchers,and its development has a good prospect.The integral imaging system is divided into two parts: recording and reproducing.In the recording part,lens array is used to image spatial scene information,and the imaging information is recorded on the recording plane.According to the principle of optical path invertibility,in the reconstruction part,the lens array with the same parameters as the recording lens array is used to image the information on the recording plane,and the reconstruction is completed by focusing and restoring the light emitted from the recording plane.Due to the inherent characteristics of integral imaging,the reconstructed image resolution obtained in the process of integral imaging reproduction is low.Therefore,this paper studies how to improve the display resolution of integral imaging.Firstly,an integral imaging method based on fast marching method and virtual viewpoint synthesis is proposed.Sparse camera array method is used for viewpoint recording,which can reduce the number of cameras in the integral imaging system and reduce the camera rendering time and complexity of the system.However,there are few viewpoints recorded by this method,which leads to the low display resolution and display quality of the integral imaging system.Therefore,this paper uses the virtual viewpoint synthesis method to increase the number of viewpoints,uses the image restoration method based on fast moving algorithm to repair the added viewpoints,and then splices the original viewpoints and synthetic viewpoints into an element image array in order,and carries out integral imaging reconstruction in the computer environment.Experimental results show that this method increases the efficiency of viewpoint acquisition and reduces the rendering cost of integral imaging system.Then,an integral imaging method based on deep learning image super-resolution and viewpoint synthesis is proposed.Firstly,an environment is built in computer software for viewpoint rendering,and virtual viewpoint synthesis is carried out with the rendered image as reference to expand viewpoint.Then,the image super-resolution network based on deep learning is used to reconstruct all viewpoints,and the reconstructed images are spliced into element image array in order.Finally,the synthesized high-resolution element image array is used for integral imaging reconstruction.Experimental results show that this method can reduce the system rendering cost and improve the display resolution of reconstruction integral imaging.