Research On Elemental Image Generation And 3D Reconstruction Based On Integral Imaging Technology

As the ultimate goal of 3D display,naked-eye stereoscopic display technology is the key direction of future research in the field of display.As a naked-eye stereoscopic display technology,integral imaging technology has the advantages of full disparity,no visual fatigue for viewers,and continuous viewpoints.It can display three-dimensional images of real scenes in multiple dimensions.It has attracted the attention of researchers from various countries since it was proposed.Due to factors such as the integral imaging display mechanism,there are still some problems to be solved.For example,the elemental image acquisition method is complex,the reconstructed light field has various crosstalks,and the reconstructed image quality is limited.This paper studies from the two main aspects of elemental image generation and 3D reconstruction based on computer-generated integral imaging,and proposes corresponding solutions for some key problems.The main innovative research work is as follows:(1)Research on integral imaging content generation method based on depth informationa fast elemental image generation algorithm is proposed for integral imaging display based on depth information and a two-dimensional scene image,which can solve the problems of complicated acquisition equipment and low quality of elemental images of real scenes.The virtual ideal light field model is constructed by the corresponding relationship between the color map and the depth information.The pixel mapping relationship between the elemental image plane and the ideal light field is established.And each pixel in the elemental image matches a unique scene pixel closest to the observer by inverse iteration along the optical path.This reverse single-point mapping makes each pixel in the elemental image correspond to only a unique scene pixel,which improves the matching accuracy,effectively eliminates holes at the depth discontinuities,and improves the display quality of the reconstructed light field.The redundant calculation of multiple selections is avoided.The time complexity of the algorithm is mainly controlled by the total number of pixels in the elemental image array,not the scene image resolution,which fundamentally reduces the number of matches.Compared with the traditional algorithm,the generation speed is increased by more than 6 times.Aiming at the defects of LED stereoscopic display,the elemental image generation algorithm that suppresses accumulated errors is proposed.To avoid the inefficiency caused by repeated calculation of scene pixels in the forward mapping algorithm,homologous point mapping is introduced.To suppress the accumulated error of homologous points and improve the reconstruction accuracy,the distance between each group of homologous points is corrected based on the actual spatial resolution limit and the displacement offset value from the reconstructed surface to the center depth plane.Fundamentally reduce the cumulative deviation in the elemental image array.the display quality of the reconstructed light field based LED is improved,and the edge distortion of the stereo image is eliminated.(2)Research on depth information accuracy for integral imaging content generationFor computer-generated integral imaging systems that utilize depth map to simplify the acquisition process,inaccurate pixel values in the depth map can lead to transitive position errors of pixels in the light field.Therefore,this paper proposes a depth information precision algorithm for integral imaging content generation from two aspects of active and passive acquisition of depth maps.The passively acquired depth map has a large number of concentrated mismatch points due to projection scaling distortion and weak texture.The accumulated depth error cause diffuse errors in the calculation of the generated elemental images.In this regard,a stereo matching algorithm based on generalized superpixel segmentation is proposed.Generalized superpixel chunk matching for weakly textured regions is first designed.The simplified linear least squares method is used to calculate the fitting degree of the custom discrete boundary,and the consistency of the discrete boundary is checked to achieve fast and high-precision boundary matching.Secondly,the adsorption growth matching is designed to deal with the remaining dense texture area.The local error range and margin of error of the depth map obtained in this way are small,and there is no patchy mismatched area,and the average mismatching rate is lower than 9.84%.The distortion and depth fracture of the reconstructed light field are eliminated,so that the reconstructed 3D image has clear edges and prominent details,and improves the integrity of the light field.In the depth map obtained by the depth camera,the front and back background holes at the edge of the object are large and the background noise points are too many.Therefore,a depth map repair algorithm for RGB-D depth camera is proposed.Combined with the improved joint bilateral filtering algorithm adding depth value similarity factor,the depth map is repaired and optimized,resulting in a highly robust depth reconstruction effect.The depth value comparison ensures that the template block on the background boundary line is preferentially selected for filling,so as to achieve a good transition from the background to the foreground area.Moreover,overfilling is avoided,the edges of the image are preserved,and the repair accuracy is high,and the average PSNR is increased from 20.91 to 36.26.(3)Research on 3D reconstruction based on human eye viewing mechanismAiming at the phenomenon of 3D reconstruction quality degradation and distortion caused by adjacent elemental images and light crosstalk between adjacent lenses in the process of integral imaging reconstruction,a 3D reconstruction quality improvement method based on the human eye viewing mechanism is proposed.First,determine the exact source of light at different viewing positions within the field of view,calculate the light distribution that actually enters the human eye in the elemental image plane,and establish a field of view model that maps the elemental image plane to a specific viewpoint.Second,the pixel information entering a specific human eye position is converted into a concatenation of linear transformations made by "fixed translation amount" and "scale factor".Finally,the optimal elemental image array corresponding to the human eye-based viewing mechanism is designed,and the pixels on the elemental image array are rearranged.In this way,the range of the viewpoint can be expanded and the anti-shake performance of the viewpoint can be improved.It can suppress the interference of the wrong light and improve the display density of the light field.Under the premise of the same reconstruction resolution,the algorithm in this paper improves the clarity of the view and increases the amount of light field information.Experiments show that the algorithm in this paper eliminates the periodic crosstalk phenomenon and expands the number of viewing viewpoints,so the viewing effect at the edge position is effectively improved.The effective viewing angle is twice as high as the traditional algorithm.The average value of the structural similarity value is greater than 0.88,which solves the problem of excessive crosstalk light in the reconstructed system.