【摘要】 针对TOF相机原始获取深度图像分辨率非常低,且超分辨率重建中易出现边缘模糊和伪影的问题,提出一种基于二阶微分算子和测地距离的深度图超分辨率重建算法。以彩色信息作为引导,运用双边滤波的思想,采用测地距离把低分辨率深度图像的空间高斯核与高分辨率彩色图像的幅度高斯核函数结合起来,体现了深度图与彩色图的一致性,并引入深度核函数对两个相邻像素具有类似颜色但深度值不同的情况进行处理,抑制颜色相似但深度值不同区域的伪影现象,恢复出边缘轮廓显著的高分辨率深度图。实验结果表明,该算法可以有效保护图像的边缘结构且解决伪影问题,并在定性和定量两个方面都可达到很好的效果。更多还原

【Abstract】 The resolution of originally captured depth image with TOF camera is very low,and it is prone to having problems of blurred edges and artifacts in super-resolution reconstruction. In light of this issue,we proposed a depth map super-resolution reconstruction method which is based on second-order differential operator and geodesic distance. The algorithm is guided with colour information,uses the idea of bilateral filtering,and combines the spatial Gaussian kernel of low resolution depth image and the magnitude Gaussian kernel function of high resolution colour image with geodesic distance,thus reflects the consistency of depth map and colour image. Moreover,the algorithm introduces the depth kernel function to deal with the situation that two adjacent pixels have similar colour but different depth values,and suppresses the artifacts in the regions with similar colour but different depth values. Then the high resolution depth map with significant edge contour is restored. Experimental results demonstrated that this algorithm can effectively preserve the edge structure of image and solve artifacts problem,and reaches very good effect in both qualitative and quantitative aspects.更多还原