光学学报, 2020, 40 (18): 1810002, 网络出版: 2020-08-27   

基于卷积神经网络去噪正则化的条纹图修复 下载: 1133次

Fringe Pattern Inpainting Based on Convolutional Neural Network Denoising Regularization
作者单位
四川大学电子信息学院光电系, 四川 成都 610065
摘要
条纹投影轮廓术测量表面存在高动态范围反射率物体时,采集的条纹图中出现的强度饱和区域将导致对应区域的相位计算误差或缺失,最终影响三维形貌的恢复。为此,提出一种基于卷积神经网络(CNN)去噪正则化的条纹图高光区域修复算法。该方法仅需要在正常曝光和短曝光条件下获取两帧条纹图,快速实现条纹修复,步骤如下:利用Otsu方法对短曝光条纹的调制度图做二值化处理以确定反光区域位置;把短曝光条纹对应区域进行灰度调节后融入正常曝光条纹中,形成迭代修复算法的初值;通过CNN去噪正则化的修复算法,实现条纹图局部高光区域的快速修复,再利用修复后的条纹实现对高动态范围反射物体的三维面形重建。与其他几种常用方法对比,所提方法在条纹修复效果和修复时间上都具有较大优势。
Abstract
Intensity saturation zone in the fringe pattern will appear when fringe projection profilometry is used to measure objects with high dynamic range reflectivity, which will affect the phase reconstruction of the tested object. In this paper, we proposed a fringe pattern inpainting method based on convolutional neural network (CNN) denoising regularization. Two fringe patterns under normal and short exposure time are respectively captured to quickly build a fringe with good quality using following steps. Otsu threshold method is used to determine highlight region by treating the modulation information of short exposure fringe pattern. Set an initial value for iteration by fusing the normal exposure fringe pattern with gray-adjusted short exposure fringe pattern. Realize fast fringe pattern inpainting using CNN denoising regularization and finally obtain a fringe to realize the high dynamic range phase reconstruction. Compared with other methods, the proposed method has advantage in effect and time of fringe inpainting.

彭广泽, 陈文静. 基于卷积神经网络去噪正则化的条纹图修复[J]. 光学学报, 2020, 40(18): 1810002. Guangze Peng, Wenjing Chen. Fringe Pattern Inpainting Based on Convolutional Neural Network Denoising Regularization[J]. Acta Optica Sinica, 2020, 40(18): 1810002.

本文已被 7 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!