摘要
受采集装置的限制,采集的深度图像存在分辨率较低、易受噪声干扰等问题.本文构建了分级特征反馈融合网络(Hierarchical feature feedback network,HFFN),以实现深度图像的超分辨率重建.该网络利用金字塔结构挖掘深度-纹理特征在不同尺度下的分层特征,构建深度-纹理的分层特征表示.为了有效利用不同尺度下的结构信息,本文设计了一种分级特征的反馈式融合策略,综合深度-纹理的边缘特征,生成重建深度图像的边缘引导信息,完成深度图像的重建过程.与对比方法相比,实验结果表明HFNN网络提升了深度图像的主、客观重建质量.
Due to the limitations of the depth acquisition devices,the depth maps are easily distorted by noise and with low resolution.In this paper,we propose a hierarchical feature feedback network(HFFN)to reconstruct depth map with high-resolution.The HFFN uses a pyramid structure to mine the features of depth-texture at different scales,and constructs the hierarchical feature representations of depth-texture.In order to effectively utilize the structural information at different scales,a feedback fusion strategy based on hierarchical features is designed,which integrates the edge features of depth-texture to generate edge guidance information to assist the reconstruction process of depth map.Compared with the comparison methods,the experimental results show that the HFFN can achieve better subjective and objective quality.
作者
张帅勇
刘美琴
姚超
林春雨
赵耀
ZHANG Shuai-Yong;LIU Mei-Qin;YAO Chao;LIN Chun-Yu;ZHAO Yao(Institute of Information Science,Beijing Jiaotong University,Beijing 100044;Beijing Key Laboratory of Modern Information Science and Network Technology,Beijing 100044;School of Computer and Communication Engineering,University of Science and Technology Beijing,Beijing 100083)
出处
《自动化学报》
EI
CAS
CSCD
北大核心
2022年第4期992-1003,共12页
Acta Automatica Sinica
基金
国家自然科学基金(61972028,61902022,U1936212)
国家重点研发计划(2018AAA0102100)
中央高校基本科研业务费(2019JBM018,FRF-TP-19-015A1)资助。
关键词
深度图像
超分辨率重建
特征融合
残差学习
Depth map
super-resolution reconstruction
feature fusion
residual learning
