Image foreground-background segmentation method based on sparse decomposition and graph Laplacian regularization
TAN Tingfang,, CAI Wanyuan, JIANG Junzheng,
1. School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2. State and Local Joint Engineering Research Center for Satellite Navigation and Location Service, Guilin University of Electronic Technology, Guilin 541004, China
3. Hangzhou Institute of Technology, Xidian University, Hangzhou 311231, China
A new method for segmenting the foreground and background of images was proposed by using the graph signal processing theory and sparse decomposition model aiming at the problem of isolated pixel points in the segmentation results of existing image foreground-background segmentation methods. The intrinsic structure of an image was modeled as a graph, and the intrinsic correlation between pixels was effectively characterized by the graph model. The pixel intensity of the image was modeled as a graph signal. The image background was linearly represented as a smooth component by a set of graph Fourier transform basis functions, the foreground overlaid on the background was a sparse component, and the connectivity between foreground pixels could be characterized by the graph Laplacian regularization term. The image foreground-background segmentation problem was reduced to a constrained optimization problem incorporating the sparse decomposition model and graph Laplacian regularization term, and the alternating direction multiplier method was adopted to solve the optimization problem. The experimental results show that the proposed method has better segmentation performance compared with other existing methods.
Keywords:graph signal processing
;
graph Laplacian regularization
;
graph Fourier transform basis function
;
sparse decomposition
;
foreground-background segmentation
TAN Tingfang, CAI Wanyuan, JIANG Junzheng. Image foreground-background segmentation method based on sparse decomposition and graph Laplacian regularization. Journal of Zhejiang University(Engineering Science)[J], 2024, 58(5): 979-987 doi:10.3785/j.issn.1008-973X.2024.05.011
将图像划分为一系列大小为$ l \times l $的非重叠图像块. 对于每个图像块,像素点建模为图节点,像素点的强度定义为图节点上的信号值,图像块中的每个像素点都与其4个邻居相连,相连的像素点之间的边的权重用来刻画像素之间的关联性. 为了捕捉相邻节点之间的强度相似性,2个相连的节点$ i $、$ j $之间的权重由高斯核计算[18,28].
假设图像块大小为$ l \times l $,基函数的数量为$ M $,向量$ {{\boldsymbol{f}}_r} $、$ {{\boldsymbol{s}}_r} $和$ {{\boldsymbol{\alpha }}_r} $的大小分别为$ {l^2} \times 1 $、$ {l^2} \times 1 $和$ M \times 1 $,矩阵$ {{\boldsymbol{P}}_r} $的大小为$ {l^2} \times M $. 由式(10)、(16)可知,第1、4个变量更新涉及矩阵乘法和逆运算. 对于大小为$ d \times d $的矩阵逆运算,计算复杂度为$ O\left( {{d^3}} \right) $,因此所提方法对于大小为$ l \times l $的图像块的完整计算复杂度为$ O\left( {{M^3}+M+{l^2}M+{l^6}} \right) $. 由于$ M $为基函数的数量,$ M < {l^2} $,所提方法对于图像块的复杂度为$ O\left( {{l^6}} \right) $.
对于给定大小为$ m \times n $的图像,假设$ m $和$ n $远大于$ l $,则大约有$ {{mn}}/{{{l^2}}} $个图像块,所提方法的整体计算复杂度为$ O\left( {mn\left( {{{{M^3}}}/{{{l^2}}}+{M}/{l^2}+M+{l^4}} \right)} \right) $. 因为$ M < {l^2} $,所提方法的整体计算复杂度为$ O\left( {mn{l^4}} \right) $.
MUKHERJEE D, CHRYSAFIS C, SAID A. JPEG2000-matched MRC compression of compound documents [C]// Proceedings. International Conference on Image Processing . Rochester: IEEE, 2002.
MINAEE S, WANG Y. Screen content image segmentation using least absolute deviation fitting [C]// IEEE International Conference on Image Processing . Quebec City: IEEE, 2015: 3295-3299.
MINAEE S, WANG Y. Screen content image segmentation using sparse decomposition and total variation minimization [C]// IEEE International Conference on Image Processing . Phoenix: IEEE, 2016: 3882-3886.
ABIKO K, URUMA K, SUGAWARA M, et al. Image segmentation based graph-cut approach to fast color image coding via graph Fourier transform [C]// IEEE Visual Communications and Image Processing . Sydney: IEEE, 2019: 457-460.
BOGACH I V, LUPIAK D D, IVANOV Y Y, et al. Analysis and experimental research of modifications of the image segmentation method using graph theory [C]// International Siberian Conference on Control and Communications . Tomsk: IEEE, 2019: 490-493.
HU W, CHEUNG G, ORTEGA A, et al
Multiresolution graph Fourier transform for compression of piecewise smooth images
THANOU D, FROSSARD P. Multi-graph learning of spectral graph dictionaries [C]// IEEE International Conference on Acoustics, Speech and Signal Processing . South Brisbane: IEEE, 2015: 3397-3401.
LIU M, WEI Y. Image denoising using graph-based frequency domain low-pass filtering [C]// IEEE 4th International Conference on Image, Vision and Computing. Xiamen: IEEE, 2019: 118-122.
ARAVKIN A, BECKER S, CEVHER V, et al. A variational approach to stable principal component pursuit [EB/OL]. [2014-06-04]. https://arxiv.org/abs/1406.1089.
Graph signal processing: overview, challenges, and applications
1
2018
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
The emerging field of signal processing on graphs: extending high-dimensional data analysis to networks and other irregular domains
1
2013
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
1
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
0
Multiresolution graph Fourier transform for compression of piecewise smooth images
0
2015
Graph Laplacian regularization for image denoising: analysis in the continuous domain
0
2017
图信号处理在高光谱图像处理领域的典型应用
1
2023
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
图信号处理在高光谱图像处理领域的典型应用
1
2023
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
Graph signal processing for machine learning: a review and new perspectives
1
2020
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
Hyperspectral image denoising using adaptive weight graph total variation regularization and low-rank matrix recovery
3
2021
... 近年来,图信号处理(graph signal processing,GSP)[9-11]作为处理非规则域数据的有效工具,广泛应用于图像处理[12-16]、计算机图形学[10]和社交网络分析[17]等领域. GSP通过图对数据的内在结构进行建模,可以有效地刻画数据样本间的关系. 在图像处理中,利用基于图的表示方法,可以捕获图像数据的内在结构特征,如分段平滑特性(piecewise smoothness, PWS)[9,18]. 对于图像数据,可以基于GSP理论将像素点建模为图节点,像素点的强度定义为图节点的信号值,相邻像素之间的相似性通过边的权重来表征,可以充分刻画图像像素之间的内在关联性. ...
... 将图像划分为一系列大小为$ l \times l $的非重叠图像块. 对于每个图像块,像素点建模为图节点,像素点的强度定义为图节点上的信号值,图像块中的每个像素点都与其4个邻居相连,相连的像素点之间的边的权重用来刻画像素之间的关联性. 为了捕捉相邻节点之间的强度相似性,2个相连的节点$ i $、$ j $之间的权重由高斯核计算[18,28]. ...
... 将图像划分为一系列大小为$ l \times l $的非重叠图像块. 对于每个图像块,像素点建模为图节点,像素点的强度定义为图节点上的信号值,图像块中的每个像素点都与其4个邻居相连,相连的像素点之间的边的权重用来刻画像素之间的关联性. 为了捕捉相邻节点之间的强度相似性,2个相连的节点$ i $、$ j $之间的权重由高斯核计算[18,28]. ...
Optimizing evaluation metrics for multitask learning via the alternating direction method of multipliers