摘要
针对主流传感器采集的深度图像存在深度信息区域缺失、噪声等图像质量问题,提出一种基于SD全局优化模型的深度图像增强算法。采用非凸函数对SD全局优化模型平滑项进行建模,使其对异常值具有较强的鲁棒性。使用基于递阶辨识(HI)的交替方向乘子法求解SD全局优化模型,将目标函数分解成多个子目标函数,并对每个子目标函数通过HI思想进行逐个求解,降低求解复杂度。实验结果表明,该算法在加快收敛速度的同时,能有效去除图像噪声及抑制深度伪影。
Because the depth image acquired by the mainstream sensor has image quality problems such as missing depth information area and noise,a depth image enhancement algorithm based on Static/Dynamic(SD) global optimization model is proposed.The SD global optimization model smoothing term is modeled by non-convex functions,which makes it more robust to outliers.The Alternating Direction Multiplier Method(ADMM) based on Hierarchical Identification(HI) is used to solve the SD global optimization model.The method decomposes the objective function into multiple sub-objective functions,and solves each sub-objective function one by one through the HI idea to reduce the complexity of the solution.Experimental results show that the proposed algorithm can effectively remove image noise and suppress depth artifacts while speeding up convergence.
引文
[1] 谭志国,欧建平,张军,等.一种层析深度图像去噪算法[J].光学学报,2017,37(5):94-100.
[2] 刘俊毅.彩色图像引导的深度图像增强[D].杭州:浙江大学,2014.
[3] 刘金荣,李淳其,欧阳建权,等.基于联合双边滤波的深度图像增强算法[J].计算机工程,2014,40(3):249-252.
[4] ZHANG Q,SHEN X,XU L,et al.Rolling guidance filter[C]//Proceedings of European Conference on Computer Vision.Berlin,Germany:Springer,2014:815-830.
[5] HE K,SUN J,TANG X.Guided image filtering[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(6):1397-1409.
[6] DIEBEL J,THRUN S.An application of Markov random fields to range sensing[EB/OL].(2006-07-11)[2018-02-17].https://arxiv.org/abs/1302.5589.
[7] GONG X J,LIU J Y,ZHOU W H,et al.Guided depth enhancement via a fast marching method[J].Image and Vision Computing,2013,31(10):695-703.
[8] HAM B,CHO M,PONCE J.Robust image filtering using joint static and dynamic guidance[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2015:4823-4831.
[9] XIAO J,NG M K P,YANG Y F.On the convergence of nonconvex minimization methods for image recovery[J].IEEE Transactions on Image Processing,2015,24(5):1587-1598.
[10] KIM Y,HAM B,OH C,et al.Structure selective depth superresolution for RGB-D cameras[J].IEEE Transactions on Image Processing,2016,25(11):5227-5238.
[11] 丁锋,杨家本.大系统的递阶辨识[J].自动化学报,1999,25(5):647-654.
[12] PARIKH N,BOYD S.Proximal algorithms[J].Founda-tions and Trends in Optimization,2013,1(3):127-239.
[13] YANG J,YE X,LI K,et al.Color-guided depth recovery from RGB-D data using an adaptive autoregressive model[J].IEEE Transactions on Image Processing,2014,23(8):3443-3458.
[14] FERSTL D,REINBACHER C,RANFTL R,et al.Image guided depth upsampling using anisotropic total generalized variation[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2013:993-1000.
[15] SONG S,LICHTENBERG S P,XIAO J A.RGB-D:a RGB-D scene understanding benchmark suite[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C.,USA:IEEE Press,2015:1-6.
[16] SHEN X,ZHOU C,XU L,et al.Mutual-structure for joint filtering[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2015:3406-3414.