CN100433046C

CN100433046C - Image blind separation based on sparse change

Info

Publication number: CN100433046C
Application number: CNB2006101166989A
Authority: CN
Inventors: 刘盛鹏; 方勇
Original assignee: University of Shanghai for Science and Technology
Current assignee: University of Shanghai for Science and Technology
Priority date: 2006-09-28
Filing date: 2006-09-28
Publication date: 2008-11-12
Anticipated expiration: 2026-09-28
Also published as: CN1936926A

Abstract

The invention relates to an image blind source separation method based on sparse transformation. This method first uses the Contourlet transform to perform multi-scale and multi-directional sparse decomposition of the received mixed image signal, and uses the sparsity criterion in the Contourlet transform domain to select the sub-image group with the best sparsity; then uses the traditional fast fixed-point The independent component analysis method performs blind separation on the selected sub-image group to obtain the separation matrix; finally, uses this separation matrix to separate the received mixed image signal, extracts each independent component in the mixed image, and achieves the goal of image blind source separation. Purpose. The image blind separation method provided by the invention can improve the accuracy of blind source separation and achieve a more ideal separation effect, and is suitable for radio communication systems, sonar and radar systems, audio and acoustics, and medical signal processing in military or non-military fields .

Description

A Blind Image Source Separation Method Based on Sparse Transformation

技术领域 technical field

本发明涉及一种图像降噪方法，特别是一种基于稀疏变换的图像盲源分离方法。在军事领域或非军事领域的图像处理中均有着重要的应用潜力。The invention relates to an image noise reduction method, in particular to an image blind source separation method based on sparse transformation. It has important application potential in image processing in military or non-military fields.

背景技术 Background technique

通常，图像在其获取或传输过程中都会受到其他信号的污染，为了后续的进一步处理，很有必要进行分离处理。图像分离的目的就是尽可能地提取接收信号中的各个独立信号分量，以提高图像的质量。目前，图像降噪方法主要分为传统的滤波方法和盲源分离方法，其中以盲源分离方法最具代表性。Usually, the image will be polluted by other signals during its acquisition or transmission process, and it is necessary to carry out separation processing for subsequent further processing. The purpose of image separation is to extract the individual signal components in the received signal as much as possible to improve the quality of the image. At present, image noise reduction methods are mainly divided into traditional filtering methods and blind source separation methods, among which blind source separation methods are the most representative.

盲源分离方法是在信源S和信号传输特征均未知的情况下，仅仅通过接收到的混合信号X来进行这些相互独立的源信号的分离。现今，主要的盲源分离方法主要有基于高阶统计量的独立分量分析方法(ICA)、基于随机梯度下降的最大熵方法(Infomax)、自然梯度学习方法(NGA)和采用负熵判据的快速ICA方法(FastICA)，即定点方法(Fixed-point)。虽然这些方法在盲源分离方面取得了较好的效果，但是，它们还不是最佳的。The blind source separation method is to separate these mutually independent source signals only through the received mixed signal X when the source S and the signal transmission characteristics are unknown. Nowadays, the main methods of blind source separation mainly include independent component analysis (ICA) based on high-order statistics, maximum entropy method (Infomax) based on stochastic gradient descent, natural gradient learning method (NGA) and Negentropy criterion. Fast ICA method (FastICA), that is, fixed-point method (Fixed-point). Although these methods achieve good results in blind source separation, however, they are not yet optimal.

研究表明，输入信号的稀疏性在很大程度上影响盲源分离方法的性能。当输入信号的越稀疏，图像盲源分离效果越好。于是，基于小波变换的图像盲源分离方法应运而生，在很大程度上提高了分离效果。但是，由一维小波通过张量积形成的二维可分离小波变换只能有效地表示一维奇异信息即点奇异信息，而不能有效地描述图像中的二维或高维奇异信息，如线、轮廓等重要信息，从而制约了基于小波变换的图像盲源分离方法的性能。轮廓小波变换Contourlet作为一种新的信号分析工具，解决了小波变换不能有效表示二维或更高维奇异性的缺点，能准确地将图像中的边缘捕获到不同尺度、不同频率、不同方向的子带中。它不仅具有小波变换的多尺度特性，还具有小波变换不具有的方向性和各向异性，因此能很好地应用于图像处理中。Studies have shown that the sparsity of input signals largely affects the performance of blind source separation methods. When the input signal is sparser, the effect of image blind source separation is better. Therefore, the image blind source separation method based on wavelet transform emerges at the historic moment, which improves the separation effect to a great extent. However, the two-dimensional separable wavelet transform formed by the one-dimensional wavelet through the tensor product can only effectively represent one-dimensional singular information, that is, point singular information, but cannot effectively describe two-dimensional or high-dimensional singular information in the image, such as line , outline and other important information, which restricts the performance of the image blind source separation method based on wavelet transform. As a new signal analysis tool, the contour wavelet transform, Contourlet, solves the shortcoming that the wavelet transform cannot effectively represent two-dimensional or higher-dimensional heterogeneity, and can accurately capture the edges in the image to different scales, different frequencies, and different directions. in the subband. It not only has the multi-scale characteristics of wavelet transform, but also has the directionality and anisotropy that wavelet transform does not have, so it can be well used in image processing.

发明内容 Contents of the invention

本发明的目的在于针对现有图像盲源分离方法方法存在的不足，提出了一种基于稀疏变换的图像盲源分离方法，该方法在轮廓小波变换Contourlet中获取分离矩阵，利用这个分离矩阵来对接收到的混合图像信号进行分离，提取混合图像中的各个独立分量，达到图像盲源分离的目的。The purpose of the present invention is to propose a kind of image blind source separation method based on sparse transformation for the deficiency that existing image blind source separation method method exists, and this method obtains separation matrix in contour wavelet transform Contourlet, utilizes this separation matrix to The received mixed image signal is separated, and each independent component in the mixed image is extracted to achieve the purpose of image blind source separation.

为了达到上述目的，本发明采用下述技术方案：In order to achieve the above object, the present invention adopts following technical scheme:

一种基于稀疏变换的图像盲源分离方法。其特征在于首先利用轮廓小波变换Contourlet对接收到的混合图像信号进行多尺度、多方向的稀疏分解，并在轮廓小波变换Contourlet域利用稀疏性判别标准来选取稀疏性最好的子图像组；然后利用传统的快速定点独立分量分析方法对选取的子图像组进行盲分离，获取分离矩阵；最后，利用这个分离矩阵来对接收到的混合图像信号进行分离，提取混合图像中的各个独立分量，达到图像盲源分离的目的。A method for blind source separation of images based on sparse transforms. It is characterized in that firstly, the received mixed image signal is subjected to multi-scale and multi-direction sparse decomposition by using the contour wavelet transform Contourlet, and the sparsity criterion is used in the contour wavelet transform Contourlet domain to select the sub-image group with the best sparsity; and then Use the traditional fast fixed-point independent component analysis method to blindly separate the selected sub-image groups to obtain the separation matrix; finally, use this separation matrix to separate the received mixed image signal and extract the independent components in the mixed image to achieve The purpose of image blind source separation.

上述图像盲源分离方法的具体步骤如下：The specific steps of the above image blind source separation method are as follows:

①初始化设置。设定轮廓小波变换Contourlet的中拉普拉斯塔式分解LP层数K和每层中的方向分解数L_k；①Initialize settings. Set the Laplastian decomposition LP layer number K and the direction decomposition number L _k in each layer of the contour wavelet transform Contourlet;

②对接收到的混合图像X1和X2分别进行多尺度、多方向的轮廓小波变换Contourlet稀疏分解，即② Perform multi-scale and multi-directional contour wavelet transform Contourlet sparse decomposition on the received mixed images X1 and X2, namely

$[[{X x 11}_{lf lf},, {X x 11}_{hf hf}^{((1,1 1,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot;,, {X x 11}_{hf hf}^{((11,, {L L}_{11}))},, {X x 11}_{hf hf}^{((2,1 2,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot;,, {X x 11}_{hf hf}^{((K K,, {L L}_{k k}))}]] = = T T ((X x 11)),, - - - - - - ((11))$

$[[{X x 22}_{lf lf},, {X x 22}_{hf hf}^{((1,1 1,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot \cdot,, {X x 22}_{hf hf}^{((11,, {L L}_{11}))},, {X x 22}_{hf hf}^{((2,1 2,1))},, \cdot &Center Dot; \cdot \cdot \cdot &Center Dot;,, {X x 22}_{hf hf}^{((K K,, {L L}_{k k}))}]] = = T T ((X x 22)),, - - - - - - ((22))$

其中T(·)为轮廓小波变换Contourlet，从而得到一幅低频子图像Xi_lf和一系列具有不同分辨率的高频子图像Xi_hf ^(k，l)，其中k∈(1，K)和l∈(1，L_k)标明子图像位于第k层拉普拉斯塔式分解LP的第l方向，i代表1或者2；where T( ) is the contour wavelet transform Contourlet, so as to obtain a low-frequency sub-image Xi _lf and a series of high-frequency sub-images Xi _hf ^{(k, l)} with different resolutions, where k∈(1, K) and l ∈(1, L _k ) indicates that the sub-image is located in the l-th direction of the k-th Laplacian decomposition LP, and i represents 1 or 2;

③根据稀疏性判断标准，选取轮廓小波变换Contourlet后的高频子图像组X1_hf ^(k，l)和X2_hf ^(k，l)中最稀疏的子图像组，记为X1_hf ^{(ksel，lsel)}和X2_hf ^{(ksel，lsel)}。本方法根据子图像组的星图分布和聚类方法来进行稀疏性判断；③ According to the sparsity judgment standard, select the most sparse sub-image group among the high-frequency sub-image groups X1 _hf ^{(k, l)} and X2 _hf ^{(k, l)} after the contour wavelet transform Contourlet, denoted as X1 _hf ^{(ksel, lsel )} and X2 _hf ^{(ksel, lsel)} . This method judges the sparsity according to the star map distribution and the clustering method of the sub-image group;

④对第③步中得到的高频子图像组X1_hf ^{(ksel，lsel)}和X2_hf ^{(ksel，lsel)}，采用传统的自然梯度学习方法NGA来进行盲源分离，获取分离矩阵W，即④ For the high-frequency sub-image groups X1 _hf ^{(ksel, lsel)} and X2 _hf ^{(ksel, lsel)} obtained in step ③, use the traditional natural gradient learning method NGA to perform blind source separation, and obtain the separation matrix W, namely

$W W = = NGA NGA (({X x 11}_{hf hf}^{((ksel ksel,, lsel lsel))},, {X x 22}_{hf hf}^{((ksel ksel,, lsel lsel))})),, - - - - - - ((33))$

其中，NGA(·)代表自然梯度学习方法NGA；Among them, NGA( ) represents the natural gradient learning method NGA;

⑤利用第④步中得到的W来分离接收到的混合信号，得到独立分量Y1和Y2有⑤ Use the W obtained in step ④ to separate the received mixed signal, and obtain the independent components Y1 and Y2 with

$[\begin{matrix} Y Y 11 \\ Y Y 22 \end{matrix}] = = W W [\begin{matrix} X x 11 \\ X x 22 \end{matrix}] . . - - - - - - ((44))$

得到的分离结果Y1和Y2即为分离出来的原信号的估计；The obtained separation results Y1 and Y2 are the estimates of the separated original signals;

上述稀疏性判据是基于选取得的子图像组的星图分布和聚类方法来进行。具体估计步骤为：The above sparsity criterion is based on the star map distribution and clustering method of the selected sub-image group. The specific estimation steps are:

(a)令 $Z_{k, l} = {[{X 1}_{hf}^{(k, l)}, {X 2}_{hf}^{(k, l)}]}^{T},$ 其中k∈(1，K)和l∈(1，L_k)；(a) order $Z_{k, l} = {[{x 1}_{hf}^{(k, l)}, {x 2}_{hf}^{(k, l)}]}^{T},$ where k ∈ (1, K) and l ∈ (1, L _k );

(b)去除信号中较小的系数分量，以消除噪声的影响；(b) remove the smaller coefficient components in the signal to eliminate the influence of noise;

(c)将所有的数据点投影到单位球面上，即Z_k，l＝Z_k，l/||Z_k，l||；(c) Project all data points onto the unit sphere, that is, Z _{k, l} = Z _{k, l} /||Z _{k, l} ||;

(d)将所有的信号点移到正半球面：如果数据点的第一个坐标 $z_{k, l}^{1} < 0,$ Z_k，l＝-Z_k，l；否则，Z_k，l＝Z_k，l；(d) Move all signal points to the positive hemisphere: if the first coordinate of the data point $z_{k, l}^{1} < 0,$ Z _{k, l} = -Z _{k, l} ; otherwise, Z _{k, l} = Z _{k, l} ;

(e)通过聚类算法来确定聚轴和聚轴中心；(e) determine poly-axis and poly-axis center by clustering algorithm;

(f)计算所有数据点到离自身最近聚轴的距离和D_k，l，并以此来衡量稀疏性，D_k，l越小，越稀疏，其星图中的聚轴就越清晰；(f) Calculate the distance and D _k,l of all data points to the nearest poly-axis to itself, and use this to measure the sparsity, the smaller and sparser D _k,l , the clearer the poly-axis in the star map;

(g)对所有的Z_k，(for k＝1，...，N_L)计算D_k，l，寻求其最小值，令 $(ksel, lsel) = \underset{k, l}{\arg \min} (D_{k, l});$ (g) Calculate D k, _l for all Z _k , (for k=1,...,N _L ), seek its minimum value, let $(ksel, lsel) = \underset{k, l}{\arg \min} ({D.}_{k, l});$

(h)因此，最稀疏的子图像组为X1_hf ^{(ksel，lsel)}和X2_hf ^{(ksel，lsel)}。(h) Therefore, the sparsest sub-image groups are X1 _hf ^{(ksel, lsel)} and X2 _hf ^{(ksel, lsel)} .

本发明方法与现有技术相比较，具有如下显而易见的突出实质性特点和显著优点：Compared with the prior art, the method of the present invention has the following obvious outstanding substantive features and significant advantages:

本发明提供的基于稀疏变换的图像盲源分离方法是首先利用轮廓小波变换Contourlet对接收到的混合图像信号进行多尺度、多方向的稀疏分解，并在轮廓小波变换Contourlet域利用稀疏性判别标准来选取稀疏性最好的子图像组；然后利用传统的快速定点独立分量分析方法对选取的子图像组进行盲分离，获取分离矩阵；最后，利用这个分离矩阵来对接收到的混合图像信号进行分离，提取混合图像中的各个独立分量，达到图像盲源分离的目的。The sparse transform-based image blind source separation method provided by the present invention is to first use the contour wavelet transform Contourlet to perform multi-scale and multi-directional sparse decomposition on the received mixed image signal, and use the sparsity criterion in the contour wavelet transform Contourlet domain Select the sub-image group with the best sparsity; then use the traditional fast fixed-point independent component analysis method to perform blind separation on the selected sub-image group to obtain the separation matrix; finally, use this separation matrix to separate the received mixed image signal , to extract each independent component in the mixed image to achieve the purpose of image blind source separation.

具体特点和优点为：The specific features and advantages are:

(1)针对现有最具有代表性的小波域阈值降噪方法中小波变换的缺点-不能有效地表示图像中的二位或高维奇异性，将轮廓小波变换Contourlet应用到图像降噪中，进行多尺度、多方向分解，为后续降噪过程提供稀疏的图像描述系数。(1) In view of the shortcomings of wavelet transform in the most representative wavelet domain threshold denoising method - it cannot effectively represent the binary or high-Vitch heterogeneity in the image, the contour wavelet transform Contourlet is applied to image denoising, Perform multi-scale and multi-directional decomposition to provide sparse image description coefficients for the subsequent noise reduction process.

(2)针对现有图像盲源分离方法方法存在的不足，提出了基于稀疏变换的图像盲源分离方法，。(2) Aiming at the deficiencies of existing image blind source separation methods, an image blind source separation method based on sparse transformation is proposed.

(3)将接收到的信号利用轮廓小波变换Contourlet进行稀疏分解，在稀疏的条件下进行盲源分离，提高图像盲源的分离精度，提高分离图像的效果。(3) The received signal is sparsely decomposed by contour wavelet transform Contourlet, and blind source separation is carried out under sparse conditions to improve the separation accuracy of image blind source and improve the effect of image separation.

(4)根据子图像组的星图分布和聚类方法来进行稀疏性判断，选取稀疏性最好的子图像组，获取准确的分离矩阵W。(4) Judging the sparsity according to the star map distribution and clustering method of the sub-image group, select the sub-image group with the best sparsity, and obtain the accurate separation matrix W.

本发明提供的基于稀疏变换的图像盲源分离方法能提高图像盲源的分离精度，达到理想的图像分离效果。在军事领域或非军事领域的无线电通信系统、声纳与雷达系统、音频与声学和医学信号处理中均有着重要的应用潜力。The image blind source separation method based on the sparse transformation provided by the invention can improve the separation accuracy of the image blind source and achieve an ideal image separation effect. It has important application potential in radio communication systems, sonar and radar systems, audio and acoustics, and medical signal processing in military or non-military fields.

附图说明 Description of drawings

图1为本发明一个实施例的基于稀疏变换的图像盲源分离方法框图。FIG. 1 is a block diagram of a sparse transformation-based image blind source separation method according to an embodiment of the present invention.

图2是图1示例盲源分离结果照片图。图中，(a)和(b)为接收到的两幅混合图像，(c)和(d)为基于小波变换的图像盲源分离方法的分离结果，(e)和(f)为采用本发明方法的分离结果。Figure 2 is a picture of the results of blind source separation shown in Figure 1. In the figure, (a) and (b) are the two mixed images received, (c) and (d) are the separation results of image blind source separation method based on wavelet transform, (e) and (f) are the separation results using this method Separation results of the inventive method.

具体实施方式 Detailed ways

本发明的一个优选实施例结合附图祥述如下：A preferred embodiment of the present invention is described as follows in conjunction with accompanying drawing:

本基于稀疏变换的图像盲源分离方法，如图1所示。首先利用轮廓小波变换Contourlet对接收到的混合图像信号进行多尺度、多方向的稀疏分解，并在轮廓小波变换Contourlet域利用稀疏性判别标准来选取稀疏性最好的子图像组；然后利用传统的快速定点独立分量分析方法对选取的子图像组进行盲分离，获取分离矩阵；最后，利用这个分离矩阵来对接收到的混合图像信号进行分离，提取混合图像中的各个独立分量，达到图像盲源分离的目的。This image blind source separation method based on sparse transformation is shown in Figure 1. First, use the contour wavelet transform Contourlet to perform multi-scale and multi-directional sparse decomposition of the received mixed image signal, and use the sparsity criterion in the contour wavelet transform Contourlet domain to select the sub-image group with the best sparsity; then use the traditional The fast fixed-point independent component analysis method blindly separates the selected sub-image group to obtain the separation matrix; finally, use this separation matrix to separate the received mixed image signal, extract each independent component in the mixed image, and achieve the image blind source purpose of separation.

具体步骤为：The specific steps are:

①初始化设置。设定轮廓小波变换Contourlet的中拉普拉斯塔式分解LP分解层数K和每层中的方向分解数L_k；①Initialize settings. Set the Laplastian decomposition LP decomposition layer K and the direction decomposition number L _k in each layer of the contour wavelet transform Contourlet;

$[[{X x 11}_{lf lf},, {X x 11}_{hf hf}^{((1,1 1,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot;,, {X x 11}_{hf hf}^{((11,, {L L}_{11}))},, {X x 11}_{hf hf}^{((2,1 2,1))},, \cdot \cdot \cdot \cdot \cdot \cdot,, {X x 11}_{hf hf}^{((K K,, {L L}_{k k}))}]] = = T T ((X x 11)),,$

$[[{X x 22}_{lf lf},, {X x 22}_{hf hf}^{((1,1 1,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot;,, {X x 22}_{hf hf}^{((11,, {L L}_{11}))},, {X x 22}_{hf hf}^{((2,1 2,1))},, \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot;,, {X x 22}_{hf hf}^{((K K,, {L L}_{k k}))}]] = = T T ((X x 22)),,$

③根据稀疏性判断标准，选取轮廓小波变换Contourlet后的高频子图像组X1_hf ^(k，l)和X2_hf ^(k，l)中最稀疏的子图像组，记为X1_hf ^{(ksel，lsel)}和X2_hf ^{(ksel，lsel)}。本方法根据子图像组的星图分布和聚类方法来进行稀疏性判断，具体方法如下：③ According to the sparsity judgment standard, select the most sparse sub-image group among the high-frequency sub-image groups X1 _hf ^{(k, l)} and X2 _hf ^{(k, l)} after the contour wavelet transform Contourlet, denoted as X1 _hf ^{(ksel, lsel )} and X2 _hf ^{(ksel, lsel)} . This method judges the sparsity according to the star map distribution and clustering method of the sub-image group. The specific method is as follows:

A.令 $Z_{k, l} = {[{X 1}_{hf}^{(k, l)}, {X 2}_{hf}^{(k, l)}]}^{T},$ 其中k∈(1，K)和l∈(1，L_k)；A. order $Z_{k, l} = {[{x 1}_{hf}^{(k, l)}, {x 2}_{hf}^{(k, l)}]}^{T},$ where k ∈ (1, K) and l ∈ (1, L _k );

B.去除信号中较小的系数分量，以消除噪声的影响；B. Remove the smaller coefficient components in the signal to eliminate the influence of noise;

C.将所有的数据点投影到单位球面上，即Z_k，l＝Z_k，l/||Z_k，l||；C. Project all data points onto the unit sphere, that is, Z _{k, l} = Z _{k, l} /||Z _{k, l} ||;

D.将所有的信号点移到正半球面：如果数据点的第一个坐标 $z_{k, l}^{1} < 0,$ Z_k，l＝-Z_k，l；否则，Z_k，l＝Z_k，l；D. Move all signal points to the positive hemisphere: if the first coordinate of the data point $z_{k, l}^{1} < 0,$ Z _{k, l} = -Z _{k, l} ; otherwise, Z _{k, l} = Z _{k, l} ;

E.通过聚类算法来确定聚轴和聚轴中心；E. determine poly-axis and poly-axis center by clustering algorithm;

F.计算所有数据点到离自身最近聚轴的距离和D_k，l，并以此来衡量稀疏性，D_k，l越小，越稀疏，其星图中的聚轴就越清晰；F. Calculate the distance and D _k,l of all data points to the nearest poly-axis to itself, and use this to measure the sparsity. The smaller and sparser D _k,l , the clearer the poly-axis in the star map;

G.对所有的Z_k，(for k＝1，...，N_L)计算D_k，l，寻求其最小值，令 $(ksel, lsel) = \underset{k, l}{\arg \min} (D_{k, l});$ G. For all Z _k , (for k=1,..., N _L ) calculate D _{k, l} , seek its minimum value, let $(ksel, lsel) = \underset{k, l}{\arg \min} ({D.}_{k, l});$

H.因此，最稀疏的子图像组为X1_hf ^{(ksel，lsel)}和X2_hf ^{(ksel，lsel)}。H. Therefore, the sparsest sub-image groups are X1 _hf ^{(ksel, lsel)} and X2 _hf ^{(ksel, lsel)} .

$W W = = NGA NGA (({X x 11}_{hf hf}^{((ksel ksel,, lsel lsel))},, {X x 22}_{hf hf}^{((ksel ksel,, lsel lsel))})),,$

$[\begin{matrix} Y Y 11 \\ Y Y 22 \end{matrix}] = = W W [\begin{matrix} X x 11 \\ X x 22 \end{matrix}] . .$

从图2可以看出，相比目前最好的基于小波变换的图像盲源分离方法，本图像盲源分离方法能更好地分离接收信号中的独立图像分量，进一步提高图像盲源的分离精度，提高分离图像的效果。It can be seen from Figure 2 that compared with the current best image blind source separation method based on wavelet transform, this image blind source separation method can better separate the independent image components in the received signal and further improve the separation accuracy of image blind sources , to improve the effect of separating images.

表1给出了本发明图像盲源分离结果的客观评价指标。Table 1 gives the objective evaluation indexes of the image blind source separation results of the present invention.

表中采用了峰值信噪比(PSNR)来衡量降噪图像的质量，进而评价本发明图像盲源分离方法的优劣。The peak signal-to-noise ratio (PSNR) is used in the table to measure the quality of the noise-reduced image, and then evaluate the advantages and disadvantages of the image blind source separation method of the present invention.

从表中也可以得出同样的结论，本图像盲源分离方法能更好地分离接收信号中的独立图像分量，进一步提高图像盲源的分离精度，提高分离图像的效果。The same conclusion can also be drawn from the table. This image blind source separation method can better separate the independent image components in the received signal, further improve the separation accuracy of the image blind source, and improve the effect of separated images.

总之，无论是从人眼视觉效果，还是从客观评价指标，均表明本发明方法具有更高的分离精度，更好的分离效果。In a word, no matter from the visual effect of human eyes or from the objective evaluation index, it shows that the method of the present invention has higher separation precision and better separation effect.

表1分离效果的客观评价指标Table 1 Objective evaluation index of separation effect

分离图像1 Separate image 1 分离图像2 Separate image 2 基于小波变换的图像盲源分离方法 Method of Blind Source Separation of Image Based on Wavelet Transform 37.2593 37.2593 58.2365 58.2365 本发明方法 The method of the present invention 48.5709 48.5709 51.0370 51.0370

Claims

1, a kind of Image Blind source separation method based on sparse conversion, it is characterized in that at first utilizing profile wavelet transformation Contourlet that the vision-mix signal that receives is carried out multiple dimensioned, multidirectional Sparse Decomposition, and utilize sparse property discrimination standard to choose the best subimage group of sparse property in profile wavelet transformation Contourlet territory; Utilize traditional quick fixed point independent component analysis method that the subimage group of choosing is carried out blind separation then, obtain separation matrix; At last, utilize this separation matrix to come the vision-mix signal that receives is separated, extract each isolated component in the vision-mix, reach the purpose that separate in the Image Blind source.

2, the Image Blind source separation method based on sparse conversion according to claim 1 is characterized in that the concrete operations step is:

1.. the initialization setting, set the direction Number of Decomposition L in the middle Laplce tower decomposition LP number of plies K of profile wavelet transformation Contourlet and every layer _k

2.. vision-mix X1 and the X2 that receives carried out multiple dimensioned, multidirectional profile wavelet transformation Contourlet Sparse Decomposition respectively, promptly

[{X 1}_{lf}, {X 1}_{hf}^{(1,1)}, \cdot \cdot \cdot, {X 1}_{hf}^{(1, L_{1})}, {X 1}_{hf}^{(2,1)}, \cdot \cdot \cdot, {X 1}_{hf}^{(K, L_{k})}] = T (X 1),

[{X 2}_{lf}, {X 2}_{hf}^{(1,1)}, \cdot \cdot \cdot, {X 2}_{hf}^{(1, L_{1})}, {X 2}_{hf}^{(2,1)}, \cdot \cdot \cdot, {X 2}_{hf}^{(K, L_{k})}] = T (X 2),

Wherein T () is profile wavelet transformation Contourlet, thereby obtains a width of cloth low frequency subgraph as Xi _LfWith a series of high frequency subimage Xi with different resolution _Hf ^{(k, l)}, wherein k ∈ (1, K) and l ∈ (1, L _k) indicating that subimage is positioned at the l direction of the tower decomposition of k layer Laplce LP, i represents 1 or 2;

3.. according to sparse property criterion, choose the high frequency subimage group X1 behind the profile wavelet transformation Contourlet _Hf ^{(k, l)}And X2 _Hf ^{(k, l)}In the most sparse subimage group, be designated as X1 _Hf ^{(ksel, lsel)}And X2 _Hf ^{(ksel, lsel)}, this method is carried out the judgement of sparse property according to the star chart distribution and the clustering method of subimage group;

4.. to the high frequency subimage group X1 that 3. obtains in the step _Hf ^{(ksel, lsel)}And X2 _Hf ^{(ksel, lsel)}, adopt traditional natural gradient learning method NGA to carry out blind source and separate, obtain separation matrix W, promptly

W = NGA ({X 1}_{hf}^{(ksel, lsel)}, {X 2}_{hf}^{(ksel, lsel)}),

Wherein, NGA () represents natural gradient learning method NGA;

5.. utilize the W that 4. obtains in the step to separate the mixed signal that receives, obtaining isolated component Y1 and Y2 has

[\begin{matrix} Y 1 \\ Y 2 \end{matrix}] = W [\begin{matrix} X 1 \\ X 2 \end{matrix}],

Separating resulting Y1 that obtains and Y2 are the estimation of the original signal of separating.

3, the Image Blind source separation method based on sparse conversion according to claim 2 is characterized in that during described step is 3. that star chart according to the subimage group distributes and step that clustering method carries out the judgement of sparse property is:

(a) order

Z_{k, l} = {[{X 1}_{hf}^{(k, l)}, {X 2}_{hf}^{(k, l)}]}^{T},

Wherein k ∈ (1, K) and l ∈ (1, L _k);

(b) remove less coefficient component in the signal, to eliminate The noise;

(c) all data points are projected on the unit sphere, i.e. Z _k, l=Z _{K, l}/ || Z _{K, l}||;

(d) all signaling points are moved on to positive hemisphere face: if first coordinate of data point

z_{k, l}^{1} < 0,

Z _{K, l}=-Z _{K, l}Otherwise, Z _{K, l}=Z _{K, l}

(e) determine poly-axle and poly-axle center by clustering algorithm;

(f) calculate all data points to gathering the distance and the D of axle recently from self _{K, l}, and weigh sparse property, D with this _{K, l}More little, sparse more, the poly-axle in its star chart is just clear more;

(g) to all Z _k, (for k=1 ..., N _L) calculating D _{K, l}, seek its minimum value, order

(ksel, lsel) = \underset{k, l}{\arg \min} (D_{k, l});

(h) therefore, the most sparse subimage group is X1 _Hf ^{(ksel, lsel)}And X2 _Hf ^{(ksel, lsel)}