Three-dimensional Denoising Of White Light Interference Signal Using Bayesian Estimation

Metallic fracture surface morphology characteristics have high values for its failure analysis, so it is of great significance to obtain metallic fracture surface data. In this paper, we proposed to use the scanning white light interferometry, which is one of effective method for 3D surface topography measurement, with simple structures, high scanning speed, and longitudinal precision up to the level of nanometers. White light interferometry measurement contains 2D spatial data acquisition and one-dimensional time series scanning process. Therefore, in the whole measurement, there would exist, simultaneously, spatial noise and time noise, which overlap on each other to form a 3D noise with certain spatial-time distribution. Further complicated degradation of weak white light interference signal would occur, which would influence effective extraction of characteristic signals later, causing serious 3D topography reconstruction error.In this paper, based on the Laplacian priori model of white light interference signal and under the condition of Gaussian transverse 2D spatial noise and longitudinal one dimensional time noise with gamma distribution, we used the Bayesian estimation and steepest descent algorithm for 3D denoising, and compared it with other denoising method. In the system construction, we use the GPU hardware acceleration and CUDA parallel algorithm programming under Visual Studio 2010 environment, realize the processing of the white light interference signal and 3D reconstruction with fast speed.The experiment demonstrated that acquisition process introduced horizontal two-dimensional space and longitudinal one-dimensional time noise. The results show that: Compared with other denoising methods, while using the three-dimensional denoising algorithm based on Bayesian estimation, the white light interference signal average peak signal-to-noise ratio would be increased by 0.8db-5.6db, average variance be decreased around 0.2-1.3. Besides, with GPU parallel acceleration, the speed of the white light interference signal processing and three-dimensional reconstruction would be increased around 9.3 times, relative to the speed when only using CPU.