颜色指导的深度图像升采样算法的对比性研究

A Comparative Study on Guided Depth Enhancement

目前深度图像升采样主要通过双边滤波、马尔科夫随机场等算法实现,这些算法针对不同特征的图像有不同的重建效果。该文对比性研究了这些算法在用于将高分辨率的深度图像与可能带噪的稀疏深度图像结合恢复高分辨率的深度图像,即利用颜色信息指导未知深度的象素点的深度重建工作上的效果。同时,该文引入了高阶马尔科夫随机场,即利用二阶平滑项来获取更准确的深度信息,建立和求解了相应的能量方程。实验表明,双边滤波普遍能获得均方值误差比较小的效果,但其细节处理上不够到位;高阶马尔科夫随机场对未知区域的深度信息的预测比马尔科夫原型及双边滤波更为准确,特别是在已知深度点比较稀疏的情况下效果更加明显。

This paper introduces and implements some typical algorithms , combining the high-resolution camera images with the low-resolution range images to up-sample the range images .We have comparative study and applicability analysis on these algorithms .Besides , we implement the high order Markov random field（ MRF） with second order smoothness term .At the same time , we build and solve the energy function accordingly .The introduction of second order smoothness term makes the prediction of depth information exacter than the traditional framework , especially when the given depth of pixels is rather sparse .The comparative study proceeds mainly on the down-sampling of the depth ground truth .