基于局部SVD的深度图像增强算法

Local SVD Algorithm for Depth Map Enhancement

针对深度图像获取过程中的分辨率低,噪声严重等缺点,提出了基于局部SVD的深度图像增强算法。本文利用高质量的彩色纹理作为先验引导条件,对低质量的深度图进行优化。该算法的主要思想是对引导图像的局部协方差矩阵进行SVD分解,根据奇异值的大小自适应增强深度图像的细节和边缘,并利用双边滤波算法进行优化。实验结果表明,该算法在增强了深度图像的同时,抑制了平坦区域的噪声,填补了边缘缺失,实现了深度图像质量的有效改善。

A depth map enhancement algorithm via local SVD was proposed to enhance the details blurred in the depth map filtered. Firstly,the color image was used as guided image to improve the depth map in guided filter. The covariance matrix obtained at each pixel was decomposed by the singular value decomposition（SVD）, then the depth map was enhanced adaptively according to the diagonal eigenvalues. Finally,optimizing the result of guided filter with Bilateral filter. Experimental results show that the proposed method performs better in enhancing the details,while reducing noise and recovering the lost region of depth map.