基于虚拟视觉的无序图像视点绘制仿真

Simulation of Unordered Image Viewpoint Rendering Based on Virtual Vision

针对传统方法PSNR值和SSIM值较低,导致图像视点绘制效果不佳的问题,研究基于虚拟视觉的无序图像视点绘制方法,以获得高质量的无序图像。利用半像素方法处理参考图像,通过正向映射得到虚拟深度无序图,采用逆向映射填补图中由多种因素形成的部分空洞,将处于虚拟视点两旁的已知视点作为参考视点,绘制虚拟视点无序图像。运用基于同态滤波的亮度校正算法校正无序图像的亮度,使用加权合成算法融合无序图像,清除图像中的大部分空洞,通过基于区域的空洞填补算法修复残留空洞,实现无序图像视点绘制。实验结果表明:以上方法所得无序图像的PSNR和SSIM值较高,能获得较理想的无序图像视点绘制效果。

Aiming at the problem that the traditional methods have low PSNR value and SSIM value, which leads to the poor effect of image viewpoint rendering, this paper studied the viewpoint rendering method of unordered image based on virtual vision to obtain high-quality unordered image. Firstly, the half-pixel method was adopted to process reference images. And then, the forward mapping was used to obtain the virtual unordered depth-map depth. Secondly, the reverse mapping was used to fill holes formed by multiple factors in images. Meanwhile, the known viewpoints on both sides of the virtual viewpoint were taken as the reference viewpoints to draw the unordered image of virtual viewpoint. Moreover, the algorithm of brightness correction based on homomorphic filtering was used to correct the brightness of unordered image. In addition, the weighted synthesis algorithm was used to fuse the unordered image and thus to remove most of the holes in the image. Finally, the hole filling algorithm based on region was adopted to repair residual holes, thus realizing the viewpoint rendering. Experimental results show that the values of PSNR and SSIM obtained by the proposed method are higher, so that a more ideal rendering effect can be obtained.