基于卷积神经网络的单帧复合图像绝对相位恢复

Absolute Phase Recovery of Single Frame Composite Image Based on Convolutional Neural Network

提出一种使用卷积神经网络从单帧复合图像中获得高质量绝对相位的方法。所提方法使用的复合图像为嵌入散斑的条纹图像,卷积神经网络包括两个子网络,分别利用复合图像中的条纹模式分量和散斑模式分量完成包裹相位的求解及展开。所提方法在相位展开的过程中使用预拍摄的平板复合图像及其条纹阶次作为辅助信息,从而保证相位展开的准确性。实验结果表明,所提方法可以利用单帧复合图像最大限度地减少投影图片的数量,能够获得较高精度的绝对相位,为实现高精度动态场景下的三维测量提供一种可行性的解决方案。

In this paper,a convolutional neural network is proposed to obtain high quality absolute phase from single frame composite images.The composite image used in the proposed method is the fringe image embedded with speckle.The convolutional neural network consists of two sub-networks,which use the fringe mode component and the speckle mode component in the composite image to solve and unfold the wrapping phase.In the process of phase unwrapping,the proposed method uses the pre-photographed composite image and its fringe order as auxiliary information to ensure the accuracy of phase unwrapping.Experimental results show that the proposed method can minimize the number of projected images by using single-frame composite images and obtain high precision absolute phase,which provides a feasible solution for 3 D measurement in high precision dynamic scenes.