基于多流对极卷积神经网络的光场图像深度估计  

DEPTH ESTIMATION OF LIGHT FIELD IMAGE BASED ON MULTI STREAM EPIPOLAR CONVOLUTIONAL NEURAL NETWORK

作　　者：王硕 王亚飞[1] Wang Shuo;Wang Yafei(School of Information and Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学信息与通信工程学院,北京100101

出　　处：《计算机应用与软件》2020年第8期194-201,共8页Computer Applications and Software

基　　金：北京市自然科学基金-海淀原始创新联合基金项目(重点研究专题)(L182039);北京市自然科学基金-海淀原始创新联合基金项目(前沿项目)(L182032);北京市自然科学基金-市教委联合资助项目(KZ201911232046)。

摘　　要：深度图的3D信息在导航、AR、三维重建等应用上发挥着重要的作用。针对现有光场图像深度估计算法精度低、速度慢的问题,提出一种基于多流对极卷积神经网络的光场深度估计方法。将光场图像进行预处理,转化为四个角度的极平面图(epipolar plane image,EPI)结构;使用光场数据增强方法来扩充训练数据量;使用神经网络对EPI数据进行特征提取,并使用两种方式进行特征融合,得到两个初始深度图;对初始深度图进行合并优化处理,得到最终的深度图。实验结果表明,该算法在均方误差、不良像素率和计算时间三个性能指标上明显优于现有算法,在光场深度估计上具有较好的准确性和泛化能力。3 D information of depth map plays an important role in navigation,AR,3 D reconstruction and other applications.Aiming at the problems of low precision and slow speed of the existing depth estimation algorithm for light field images,we propose a depth estimation method of light field based on multi stream epipolar convolutional neural network(MS-EPINET).The light field image was preprocessed and converted into an epipolar plane image(EPI)structure with four angles;the proposed light field data enhancement method was used to augment the training data;the neural network was used to extract features from EPI data,and the feature fusion was performed in two ways to obtain two initial depth maps;we merged and optimized the initial depth map to obtain the final depth map.The experimental results show that our algorithm is superior to the existing algorithm in terms of mean square error,bad pixel rate and calculation time,and it has better accuracy and generalization ability in light field depth estimation.

关 键 词：机器视觉 光场 深度信息估计 多流对极卷积神经网络 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]