학술논문
基于灵活 LBP 纹理字典构造及多特征描述的改进 SCSR 算法 / Improved SCSR Algorithm on the Basis of Flexible LBP Texture Dictionary and Multi-Feature Description
Document Type
Academic Journal
Author
Source
华南理工大学学报（自然科学版） / Journal of South China University of Technology (Natural Science Edition). (3):57-65
Subject
Language
Chinese
ISSN
1000-565X
Abstract
针对超分辨率重构字典对结构区分度不够、在最优匹配原子搜索中耗时太长的问题，提出了一种多特征联合的分级字典（MFJD）．首先，分别用边缘块梯度特征和纹理块局部二值模式（LBP）特征来构建两种分类字典，用于逼近不同类型结构；其次，采用树结构来聚类原子，实现同一字典下的快速原子匹配；最后，引入双边总变分（BTV）正则项来约束重构结果．实验表明：与经典稀疏编码超分辨率重构（SCSR）算法相比，MFJD 多特征联合的分级字典使重构图像的 PSNR 值提高了0．2424 dB，使平均结构相似度（MSSIM）和特征相似度（FSIM）分别提高了0．0043和0．0056；由于结构分类字典维数降低，重构时间降至 SCSR 算法的22．77％．
A multi-feature joint dictionary (MFJD)is suggested to improve the structural distinction in dictionary training and to accelerate the atom matching in sparse reconstruction.Firstly,two dictionaries branched respectively for edge-and texture-descriptions are created using gradient and LBP operators.Secondly,tree structures are intro-duced to represent the hierarchical clustering of atoms,which leads to a quick atom searching.Then,bilateral total variation (BTV)regularization is employed to achieve the optimal resolution.Experimental results show that,in comparison with the sparse coding super-resolution reconstruction (SCSR)algorithm,MFJD averagely improves the PSNR,MSSIMand FSIMof an image by 0.2424 dB,0.0043 and 0.005 6,respectively,and reduces the recon-struction time to approximately 22.77% of that of SCSR algorithm owing to the reduction of dictionary dimensionality.
A multi-feature joint dictionary (MFJD)is suggested to improve the structural distinction in dictionary training and to accelerate the atom matching in sparse reconstruction.Firstly,two dictionaries branched respectively for edge-and texture-descriptions are created using gradient and LBP operators.Secondly,tree structures are intro-duced to represent the hierarchical clustering of atoms,which leads to a quick atom searching.Then,bilateral total variation (BTV)regularization is employed to achieve the optimal resolution.Experimental results show that,in comparison with the sparse coding super-resolution reconstruction (SCSR)algorithm,MFJD averagely improves the PSNR,MSSIMand FSIMof an image by 0.2424 dB,0.0043 and 0.005 6,respectively,and reduces the recon-struction time to approximately 22.77% of that of SCSR algorithm owing to the reduction of dictionary dimensionality.