Research On Light Field Information Processing Technology In Three-dimensional Display

Traditional two-dimensional display device can not provide the depth information of scene, in order to meet the urgent need of observing three-dimensional(3 D) data, it is necessary to provide a 3D display which is accustomed to human vision system. Nowadays, go with the development of display device and computer technology, 3D display industry ushers in its best times. Instead of stereo display,autostereoscopic display is achieved with denser and smoother viewpoints, larger angle of view and higher resolution. So display contents also change from two parallax images to dense light field. As a result, the amount of the data of the display contents become large and data structure of the display contents is different. In conclusion, it is challenging for content acquisition and image processing to provide display contents for autostereoscopic display. By taking advantage of the new features and new structure of dense light field, this paper studies light field processing methods which fit for dense light field and realizes kinds of applications on 3D display.Recently, the image processing method for autostereoscopic display still follows the binocular case, which does not consider the light field as a whole and do not take advantage of the statistics of light field trajectory. In consideration of above problems,light field information processing technology in three dimensional display are researched in this paper.The research contents and innovational points of this paper are as follow:(1) Depth-tunable dense viewpoint three-dimensional display based on light field control.In three-dimensional display,the parameters of acquisition equipments and display equipments usually mismatch and it leads to an imbalance or even wrong stereoscopic sense. Thus parallax adjustment is needed before display. The hardware based parallax adjustment make the acquisition equipment complex and limited the generality of 3D content. The traditional binocular stereoscopic parallax adjustment methods do not consider the statistical relationship among multi-view images. In those methods each image pair deals separately,so parallax estimation results are jitter among views, and the jitter leads to an unsmooth motion parallax and also the computational amount is large. Thus it makes sense to propose a parallax adjustment method which is suitable for dense viewpoint 3D display. In this paper, multi-view images captured with parallel camera array are considered as a 3D light field,take advantage of the statistics of linear light field trajectory the scene depth can be obtained via linear regression. Then according to the obtained depth, light field shearing is conducted to adjusting the parallax globally.(2) Depth-reversal-free three-dimensional display by adjusting the light field alignmentThe depth-reversal effect which significantly reduces 3D effect and viewing quality is considered as an inherent disadvantage of the lenticular based and the parallax barrier based autostereoscopic display. Traditional methods track viewers'position with additional hardware to avoid viewers fall into the depth reversal area but it is complex in hardware structure and do not meet the demand of viewing with multi-viewers. In this paper,the depth reversal effect is eliminated by aligning the displayed light field. Depth-reversal-free conditions are carried out and displayed light field editing theory is given. At the same time, a multi-viewing-zone integration method is proposed to make it possible that the viewpoint number of a displayed light field can be larger than the viewpoints number of a display.(3) Augmented reality three-dimensional display with light field fusionRecently, augmented reality, virtual reality and other computational display technology growths rapidly. However, as a result of unavailable light field depth information and the lack of image fusion method to keep a consistent funsion effect among light field, there is still no augmented reality technology for dense viewpoint 3D display. In this paper, real scene and virtual object are both considered as light field,and a consistent funsion effect can be realized by ligh field fusion. As the light field trajectories are linear and different objects at different depth corresponding to lines with different slope, the occlusion can be solved by comparing the "lines' slope".At the same time,an edge segmentation method with light field trajectory referenced region redistribution is proposed, thus an accurate occlusion is realizes. Finally,binocular and monocular depth cues are consistent in 3D display.(4) Profile preferentially partial occlusion removal for computional integral imagingWith the development of computer technology, computational imaging become vitality. Compared with traditional imaging methods in optical ways , the computational imaging can break through the physical limitation of optical device and realizes ideal imaging quality. Thus the computational imaging technology is often used to assist HVS to perceive a scene that can not be seen by HVS directly. In this paper, a partial occlusion removal method for integral imaging is proposed by combining light field theory and computational imaging together. The dense viewpoint camera array captures various perspectives and different portions of background can be recorded in different perspectives without occlusion, thus a clear complete background can be reconstructed with the help of computational integrated imaging which filters light rays emit from occlusion perfectly. Compared with traditional optical integrated imaging, computational integrated imaging method greatly enhances the reconstruction PSNR. In our method, significant contrast at the profile of occlusion is taken into consideration for accurate depth estimation and the profile is determined preferentially,then depth information is spread all over the light field to determine the entire occlusion, finally computational reconstruction filters the light rays emit from occlusion to realize a clear complete background. The profile preferentially occlusion decision makes the occlusion profile details accurate.Moreover, when determine the background, the structure and texture of occlusion are not considered so the method is robust to occlusion with different features. Compared with previous computational integrated imaging method the accuracy and generality of occlusion removal are enhanced.In this paper,according to different applications of three-dimensional displays,different light field control methods are applied. When dealing with multi-view content, the light field theory is taken into consideration which improves the multi-images processing effect. Different applications based on three-dimensional display are researched in our works and our works bridges three-dimensional acquisition and three-dimensional display which makes significant sense in the establishment of the industry chain of three-dimensional display.