一种基于时间导向的DIBR空洞填补方法

在立体成像系统中,填充新视点合成时图像中产生的空洞是虚拟视点合成技术的重要环节,已有方法是对空洞区域周围的像素进行水平的推测运算来得到填洞像素,其正确率不高而且容易造成图像边缘失真。提出一种基于时间导向的DIBR空洞填补方法,首先通过结构相似帧的方法得到与关键帧拥有相同背景的前后帧图像,然后比较空洞位置对应的前后帧图像的深度值,利用背景信息对空洞区域进行填补。实验结果表明,新方法得到的填补后的图像效果可以较为真实地还原特定的场景,与原始视点的结构相似性可达到91.36%,比传统方法提高了8.38%。

引入零视差缓冲的DIBR改良方法

针对图像部分像素深度值失真,导致图像变形后物体边缘失真的问题,提出了一种引入"零视差缓冲"的方法。首先,选取4幅具有不同复杂程度场景的图像,引入零视差缓冲对图像进行变形;其次采用传统图像质量评价体系(结构相似度SSIM和模糊系数K)对得到的图像进行质量评价;最后,对比传统DIBR方法和引入零视差缓冲方法得到的视点图中物体边缘的表现。实验结果表明,该方法对于变形图像的SSIM平均提高了4.66%,K平均下降了1.73%。提高了视点图中物体边缘的质量,提高了3D图像的整体质量。

Compact auto-stereoscopic display based on directional backlight using side-glowing polymer optical fiber array

A compact directional backlight module of time-multiplexed auto-stereoscopic display based on side-glowing polymer optical fiber(SGPOF)is proposed.The optical system is mainly composed of SGPOF array and cylindrical lens array.Spatial crosstalk is reduced by inserting a grating film as multi-slit diaphragm between the SGPOF array and the cylindrical lens array.A theoretical model is constructed based on the imaging optics principle of the off-axis ray.In the experiments,the cylindrical lens array concentrates a small number of views on three different view zones,the display can provide high luminance.The measurement results show that the luminance uniformity of the backlight module is up to 89.6%,and in the viewing zone the crosstalk is lower than 10%.The backlight module is compacted that the thickness being only 7 mm.The full-resolution and low-crosstalk 3D images are realized by using SGPOF backlight.