Research On Depth Estimation Method Of Light Field Image

In the display technology of light field images,integrated imaging display has shown its broad development prospects.Today,with the rapid development of science and technology,people are no longer satisfied with the two-dimensional display,and instead turn their attention to the threedimensional display,so the true three-dimensional integrated imaging display technology has gradually attracted people's attention.How to collect and three-dimensionally reconstruct the realworld objects we want,and how to display them truly,has become a key concern for researchers.The content generation part of integrated imaging contains many important steps,among which three-dimensional reconstruction is the key,and the focus of three-dimensional reconstruction is the depth estimation of the image.This article focuses on how to improve the quality and effect of the depth map.This work is of great significance to the entire 3D reconstruction and even the integrated imaging work.The research contents of this paper include:1.A multi-viewpoint input superpixel regularized light field image depth estimation method is proposed.Since the key problem of depth estimation is to deal with the edges of mutual occlusion of objects,this chapter first analyzes the generation of partially occluded boundary regions and its uncertainty,and then introduces the algorithm to solve the problem of partially occluded boundaries—SLIC superpixel algorithm.Finally,the multi-viewpoint image previously taken in the laboratory was used instead of a single image as the experimental input,and several multiviewpoint light field images in the light field image data set were used as input.After a series of algorithm processes,a depth map was obtained,and a single image input Comparing the obtained depth maps,the depth maps obtained in this experiment reflect more complete depth information,and the depth information of the object occlusion boundary part is more accurate and accurate.2.In order to improve the image resolution in depth estimation,a method of simultaneous superresolution reconstruction and light field image depth estimation is proposed.Use the SRCNN super-resolution algorithm to super-resolution reconstruction of the two intermediate images in the depth estimation process to achieve the purpose of improving the resolution of the final output image.First,we used the SRCNN algorithm on the depth estimation map t(x)of input picture pixels.Then,we used the algorithm in the confidence shrinking process of the occlusion boundary area,so that the super-resolution reconstruction work and the image depth estimation are performed simultaneously.3.Quantitative analysis of depth estimation experiment was carried out,and the PSNR and RMSE of the experimental results were calculated.Experimental results show that the PSNR of the reconstructed image is around 40 d B,and the RMSE between the obtained depth map and the standard depth map is reduced by 0.53% on average.After two algorithmic processing,after the final depth map is enlarged,a significant improvement in the detail information can be observed on the senses,and the peak signal-to-noise ratio of the objective evaluation index is also significantly improved.