CN101404084A

CN101404084A - Infrared image background suppression method based on Wavelet and Curvelet conversion

Info

Publication number: CN101404084A
Application number: CNA2008101375339A
Authority: CN
Inventors: 谷延锋; 郭琰; 刘星; 韩景龙; 张晔
Original assignee: Harbin Institute of Technology Shenzhen
Current assignee: Harbin Institute of Technology Shenzhen
Priority date: 2008-11-14
Filing date: 2008-11-14
Publication date: 2009-04-08

Abstract

Infrared image background suppression method based on Wavelet and Curvelet transformation, which involves an infrared image processing method, to solve the problem of suppressing background interference in the field of infrared image processing, while achieving background clutter suppression, it will also suppress Questions of useful target information. The original infrared image is decomposed by Wavelet transform and Curvelet transform respectively; the high-frequency decomposition coefficients at different resolutions are reserved for the Wavelet coefficients and the Curvelet coefficients respectively, and the low-frequency decomposition coefficients are set to zero; the decomposition is performed by the retained Wavelet coefficients and Curvelet coefficients respectively Image reconstruction is performed on the original infrared image; through the multi-resolution decomposition framework, image fusion is performed on the Wavelet coefficient reconstructed image and the Curvelet coefficient reconstructed image obtained in step 4, respectively, and the final infrared image background suppression result is obtained.

Description

Background Suppression Method of Infrared Image Based on Wavelet and Curvelet Transform

技术领域 technical field

本发明涉及一种红外图像处理方法，属于图像处理领域。The invention relates to an infrared image processing method, which belongs to the field of image processing.

背景技术 Background technique

一个红外自动目标识别系统一般由目标检测、目标识别、目标捕获、目标高精度跟踪和攻击点选择等功能模块组成。目标检测作为红外成像制导系统中最前端的处理环节，是精确制导中的一个关键性问题。为了尽可能早地发现目标，使制导系统有足够的反应时间并提高防御武器的预警距离，要求在很远的距离上就能够检测到目标，从而可以尽早发现目标。当探测距离和成像视场增大时，即使目标本身很大，在成像平面内也仅表现为几个像素，甚至不到一个像素，称为小目标。此时，可检测信号相对较弱，特别是在非平稳的起伏背景干扰下，目标甚至被大量复杂的噪声(杂波)所淹没，图像信噪比极低，使点目标检测工作变得很困难。因此，为了突出小目标，提高信噪比，从而提高目标检测概率，对红外小目标图像进行检测前的背景抑制和噪声滤除是十分必要的。在自动目标识别系统中，背景抑制和噪声削减统称为检测前滤波预处理。红外图像背景抑制的最终目的是为了消除杂波干扰、保存目标信息。An infrared automatic target recognition system generally consists of functional modules such as target detection, target recognition, target acquisition, target high-precision tracking, and attack point selection. As the front-end processing link in the infrared imaging guidance system, target detection is a key issue in precision guidance. In order to find the target as early as possible, so that the guidance system has enough reaction time and improve the early warning distance of defensive weapons, it is required to detect the target at a very long distance, so that the target can be found as early as possible. When the detection distance and imaging field of view increase, even if the target itself is large, it will only appear as a few pixels or even less than one pixel in the imaging plane, which is called a small target. At this time, the detectable signal is relatively weak, especially under the interference of non-stationary undulating background, the target is even submerged by a large amount of complex noise (clutter), and the image signal-to-noise ratio is extremely low, which makes the point target detection work very difficult. difficulty. Therefore, in order to highlight small targets, improve the signal-to-noise ratio, and thus increase the probability of target detection, it is necessary to perform background suppression and noise filtering on infrared small target images before detection. In an automatic target recognition system, background suppression and noise reduction are collectively referred to as filtering preprocessing before detection. The ultimate goal of infrared image background suppression is to eliminate clutter interference and preserve target information.

目前在红外图像处理领域内对背景干扰进行抑制的方法大致可以分为两大类：时-空域处理和频域/变换域处理。前者的典型方法包括线性背景预测、Top-hat变换等，它们通过空域滤波来实现红外图像背景的预测，利用预测图和原始图做差得到背景抑制的结果；后者的典型方法包括高通滤波、Wavelet变换等，主要是通过对红外图像频率成分的分析或者通过多分辨率分解，去除或抑制包含背景信息的频率分量或者变换系数，从而达到抑制背景的目的。在上述方法中，Wavelet变换以其具有的信号多尺度分解与描述能力，在红外图像背景抑制中表现出较好的性能。但Wavelet变换仅适于描述具有零维奇异性的信号，也就是说，Wavelet变换在实现背景杂波抑制的同时，也会抑制掉有用的目标信息，特别是线特征(如道路、桥梁，机场、港口等区域的边缘)。At present, the methods for suppressing background interference in the field of infrared image processing can be roughly divided into two categories: time-space domain processing and frequency domain/transform domain processing. The typical methods of the former include linear background prediction, Top-hat transformation, etc. They realize the prediction of the infrared image background through spatial filtering, and use the difference between the predicted image and the original image to obtain the result of background suppression; the typical methods of the latter include high-pass filtering, Wavelet transform, etc., mainly through the analysis of the frequency components of infrared images or through multi-resolution decomposition, remove or suppress the frequency components or transformation coefficients containing background information, so as to achieve the purpose of suppressing the background. Among the methods mentioned above, Wavelet transform shows better performance in infrared image background suppression because of its multi-scale signal decomposition and description capabilities. However, the Wavelet transform is only suitable for describing signals with zero-wich anisotropy. That is to say, the Wavelet transform will suppress useful target information while achieving background clutter suppression, especially line features (such as roads, bridges, airports, etc.) , ports, etc.).

发明内容 Contents of the invention

本发明为解决在红外图像处理领域内对背景干扰进行抑制的方法存在的在实现背景杂波抑制的同时也会抑制掉有用的目标信息的问题，提供一种基于Wavelet和Curvelet变换的红外图像背景抑制方法。本发明包括以下步骤：In order to solve the problem in the method of suppressing background interference in the field of infrared image processing that suppresses useful target information while realizing background clutter suppression, the present invention provides an infrared image background based on Wavelet and Curvelet transformation suppression method. The present invention comprises the following steps:

步骤一、对原始红外图像分别进行Wavelet变换分解和Curvelet变换分解；Step 1, performing Wavelet transform decomposition and Curvelet transform decomposition on the original infrared image respectively;

步骤二、对步骤一中分解得到的Wavelet系数，保留不同分辨率下的高频分解系数，将低频分解系数置零；Step 2. For the Wavelet coefficients decomposed in step 1, retain the high-frequency decomposition coefficients at different resolutions, and set the low-frequency decomposition coefficients to zero;

步骤三、对步骤一中分解得到的Curvelet系数，保留不同分辨率下的高频分解系数，将低频分解系数置零；Step 3. For the Curvelet coefficients decomposed in step 1, retain the high-frequency decomposition coefficients at different resolutions, and set the low-frequency decomposition coefficients to zero;

步骤四、分别通过保留的Wavelet系数和Curvelet系数对分解的原始红外图像进行图像重构；Step 4, performing image reconstruction on the decomposed original infrared image through the retained Wavelet coefficients and Curvelet coefficients respectively;

步骤五、通过多分辨率分解框架，对步骤四所得到的Wavelet系数重构图像和Curvelet系数重构图像分别进行图像融合，得到最终的红外图像背景抑制结果。Step 5. Through the multi-resolution decomposition framework, image fusion is performed on the Wavelet coefficient reconstructed image and the Curvelet coefficient reconstructed image obtained in Step 4, respectively, to obtain the final infrared image background suppression result.

本发明的工作原理是：分别利用Wavelet和Curvelet对输入原始红外图像进行背景抑制，在背景抑制结果中，Wavelet处理结果图像中能够在抑制背景的同时，较好地保存目标点特征信息，Curvelet处理结果图像中能够较好地保存目标线特征信息。对Wavelet和Curvelet背景抑制后的图像，进行基于Wavelet多分辨率分析和局部区域图像熵准则的特征融合，得到有效抑制背景同时较好地保存了目标点特征信息和线特征信息的结果图像。为了充分发挥Wavelet和Curvelet信号描述能力，本发明中采用了非抽样Wavelet变换(Undecimated Wavelet Transform，UWT)和第二代Curvelet变换(快速离散Curvelet变换，Fast Discrete Curvelet Transform，FDCT)。The working principle of the present invention is: respectively use Wavelet and Curvelet to carry out background suppression on the input original infrared image, in the background suppression result, the Wavelet processing result image can better preserve the target point feature information while suppressing the background, Curvelet processing As a result, the feature information of the target line can be well preserved in the image. For the image after Wavelet and Curvelet background suppression, the feature fusion based on Wavelet multi-resolution analysis and local area image entropy criterion is carried out to obtain the result image that effectively suppresses the background and preserves the target point feature information and line feature information. In order to give full play to Wavelet and Curvelet signal description capabilities, the present invention adopts Undecimated Wavelet Transform (Undecimated Wavelet Transform, UWT) and second-generation Curvelet Transform (Fast Discrete Curvelet Transform, Fast Discrete Curvelet Transform, FDCT).

本发明与现有技术相比，具有如下优点：Compared with the prior art, the present invention has the following advantages:

(1)突破了现有基于Wavelet变换的红外图像背景抑制方法存在的不足，即此类方法仅能较好地抑制背景，但同时也损失了目标信息。而本发明方法既能有效地抑制背景，又能很好地保存目标信息，这样才达到了红外图像背景抑制的真正目的。(1) It breaks through the shortcomings of existing infrared image background suppression methods based on Wavelet transform, that is, such methods can only suppress the background well, but at the same time lose target information. However, the method of the present invention can not only effectively suppress the background, but also preserve the target information well, so that the real purpose of infrared image background suppression is achieved.

(2)在目标信息保存方面，本发明方法能够在有效地保存目标信息的基础上，更进一步地保证目标信息的多样性，即既能够有效地保存目标的点特征信息，又能够有效地保存目标的线特征信息。点特征信息和线特征信息的保存是红外图像背景抑制的技术难点，也是后续红外图像目标检测与识别的关键。(2) In terms of object information preservation, the method of the present invention can further ensure the diversity of object information on the basis of effectively preserving object information, that is, it can not only effectively preserve the point feature information of the object, but also effectively preserve Line feature information of the target. The preservation of point feature information and line feature information is a technical difficulty in infrared image background suppression, and it is also the key to subsequent infrared image target detection and recognition.

(3)由于同时利用了Wavelet和Curvelet对图像的多尺度描述特性，抑制背景效果良好，可大大提高如红外目标检测与识别、火场红外图像监测等系统的目标探测能力和定位精度。(3) Since the multi-scale description characteristics of Wavelet and Curvelet are used at the same time, the effect of suppressing the background is good, which can greatly improve the target detection ability and positioning accuracy of systems such as infrared target detection and recognition, and fire infrared image monitoring.

附图说明 Description of drawings

图1为本发明的图像处理流程图；图2为本发明的工作流程图；图3为本发明所采用的非抽样Wavelet变换与传统抽样Wavelet变换对一维信号分解的对比：传统抽样Wavelet变换与非抽样Wavelet变换对一维信号进行三层分解的对比图，其中(a)为传统DWT分解示意图，(b)为非抽样Wavelet(UWT)分解示意图；图4为本发明所采用的快速离散Curvelet变换的频率空间区域分块图。Fig. 1 is the image processing flow chart of the present invention; Fig. 2 is the working flow chart of the present invention; Fig. 3 is the non-decimation Wavelet transformation adopted in the present invention and the contrast of traditional sampling Wavelet transformation to one-dimensional signal decomposition: traditional sampling Wavelet transformation A comparison diagram of three-layer decomposition of one-dimensional signal with non-sampled Wavelet transform, wherein (a) is a schematic diagram of traditional DWT decomposition, and (b) is a schematic diagram of unsampled Wavelet (UWT) decomposition; Fig. 4 is the fast discrete method adopted by the present invention Block diagram of the frequency-space region of the Curvelet transform.

具体实施方式 Detailed ways

具体实施方式一：参见图1～图4，本实施方式由以下步骤组成：Specific implementation mode 1: Referring to Figures 1 to 4, this implementation mode consists of the following steps:

步骤一：对原始红外图像分别进行Wavelet变换分解和Curvelet分解。Step 1: Perform Wavelet transform decomposition and Curvelet decomposition on the original infrared image respectively.

首先，对输入的原始红外图像进行非抽样Wavelet变换(UndecimatedWavelet Transform，UWT)分解。First, undecimated Wavelet Transform (UWT) decomposition is performed on the input original infrared image.

传统的离散小波变换在分解信号时采用的是抽样操作，这使得原始信号的部分时域特征不能保留在分解结果中，并且分解结果是平移可变的。为了克服这一缺陷并得到更完整的被分析信号的特性，本发明采用了非抽样小波变换，其基本原理是不对信号进行抽样，因此它是平移不变的，并且提供了丰富的时域特征信息和精确的频率局部化信息。The traditional discrete wavelet transform adopts sampling operation when decomposing the signal, which makes part of the time-domain characteristics of the original signal cannot be preserved in the decomposition result, and the decomposition result is variable in translation. In order to overcome this defect and obtain more complete characteristics of the analyzed signal, the present invention adopts non-decimated wavelet transform, the basic principle of which is not to sample the signal, so it is translation invariant and provides rich time-domain features information and precise frequency localization information.

非抽样小波变换(Undecimated Wavelet Transform，UWT)是一种无抽取离散小波变换方法，其在变换中没有进行下采样操作，具有各向同性和平移不变性。具体操作是利用高、低通滤波器(h，g)分别对图像进行滤波处理，每一级处理之后便得到变换后的高低频系数，然后对滤波器组(h，g)分别进行隔行隔列上采样，具体的采样方法是在每一级分解时，对上一级滤波器进行隔行、隔列插0处理，使得其滤波频率范围扩大一倍。这样通过UWT变换的图像就可以得到一系列的高频分量和低频分量，即S＝(S¹，S²，…，S^J，A^J)，其中，S^J是分解到第j层的高频系数，A_J是分解J层的低频系数。其重构方法如式(1)所示：Undecimated Wavelet Transform (UWT) is a non-decimation discrete wavelet transform method, which does not perform down-sampling operation in the transform, and has isotropy and translation invariance. The specific operation is to use the high-pass and low-pass filters (h, g) to filter the image respectively, and after each stage of processing, the transformed high and low frequency coefficients are obtained, and then the filter bank (h, g) is interlaced Up-sampling by column, the specific sampling method is to interleave the upper-level filter and insert 0 every other column when each level is decomposed, so that the filtering frequency range is doubled. In this way, a series of high-frequency components and low-frequency components can be obtained from the image transformed by UWT, that is, S=(S ¹ , S ² ,..., S ^J , A ^J ), where S ^J is the high-frequency component decomposed into the jth layer A _J is the low-frequency coefficient of the decomposed J layer. Its reconstruction method is shown in formula (1):

$R R = = {Σ Σ}_{j j = = 00}^{J J} \frac{11}{22} [[(({h h}^{j j} * * {A A}_{j j + + 11})) + + (({g g}^{j j} * * {S S}_{j j + + 11}))]] - - - - - - ((11))$

其中(h，g)分别为重构高低通滤波器，为了能够实现完备的图像重构，其在频域中与分解滤波器必须满足式(2)关系：Where (h, g) are reconstructed high and low pass filters respectively. In order to achieve complete image reconstruction, the relationship between them and the decomposition filter in the frequency domain must satisfy the formula (2):

$H h (({z z}^{- - 11})) \overset{~ ~}{H h} ((z z)) + + G G (({z z}^{- - 11})) \overset{~ ~}{G G} ((z z)) = = 11 - - - - - - ((22))$

其中，H、G分别是分解滤波器的频域系数，

分别是重构滤波器的频域系数。Among them, H and G are the frequency domain coefficients of the decomposition filter respectively,

are the frequency-domain coefficients of the reconstruction filter, respectively.

其次，对输入的原始红外图像进行快速离散Curvelet分解。Second, a fast discrete Curvelet decomposition is performed on the input raw infrared image.

在对图像进行Curvelet分解时，本发明采用的是快速离散Curvelet变换(Fast Discrete Curvelet Transform，FDCT)，即二代Curvelet变换。FDCT是在原有的Curvelet变换基础上发展起来的二代Curvelet变换，其原理上简化，计算更快速。一代Curvelet变换是将图像进行子带分解后不同尺度子带图像采用不同大小的分块，对每个块进行Ridgelet分解得到Curvelet系数，其变换构造中有7个参数，导致变换的数学分析比较复杂，且采用的层叠窗口实现加大了变换实现的难度。FDCT将参数减少到3个，采用更简单、更透明的结构来构造Curvelet变换，实现过程中无需用到Ridgelet分解，并且采用快速傅立叶变换加速，大大简化了Curvelet变换过程。When performing Curvelet decomposition on an image, what the present invention adopts is fast discrete Curvelet transform (Fast Discrete Curvelet Transform, FDCT), namely the second-generation Curvelet transform. FDCT is a second-generation Curvelet transform developed on the basis of the original Curvelet transform, which is simplified in principle and faster in calculation. The first-generation Curvelet transform is to decompose the image into sub-bands and use sub-blocks of different sizes for sub-band images of different scales. Ridgelet decomposition is performed on each block to obtain Curvelet coefficients. There are 7 parameters in the transformation structure, which makes the mathematical analysis of the transformation more complicated. , and the implementation of cascaded windows increases the difficulty of transformation implementation. FDCT reduces the parameters to 3, uses a simpler and more transparent structure to construct the Curvelet transform, does not need to use Ridgelet decomposition in the implementation process, and uses fast Fourier transform to accelerate, which greatly simplifies the Curvelet transform process.

在Curvelet变换的基础上，FDCT可如下表述：On the basis of Curvelet transform, FDCT can be expressed as follows:

式中

是离散Curvelet(上标D表示离散的)，式(3)是仿照式连续Curvelet变换形式。In the formula

is a discrete Curvelet (the superscript D means discrete), and Equation (3) is a continuous Curvelet transformation form modeled on the formula.

FDCT采用同中心的方块区域

来代替，以适合图像在离散化处理时在二维笛卡尔坐标系下的处理，其分块方式如图4显示。FDCT uses concentric square regions

Instead, in order to be suitable for the processing of the image in the two-dimensional Cartesian coordinate system during the discretization process, the block method is shown in Figure 4.

FDCT定义的笛卡尔坐标系下的局部窗为：The local window in the Cartesian coordinate system defined by FDCT is:

${\overset{~ ~}{U u}}_{j j} ((w w)) = = {\overset{~ ~}{W W}}_{j j} ((w w)) {V V}_{j j} ((w w)) - - - - - - ((44))$

其中in

$\{\begin{matrix} {\overset{~ ~}{W W}}_{j j} ((w w)) = = \sqrt{{Φ Φ}_{j j + + 11}^{22} ((w w)) - - {Φ Φ}_{j j}^{22} ((w w))},, j j &GreaterEqual; &Greater Equal; 11 \\ {V V}_{j j} ((w w)) = = V V ((22^{[[- - j j / / 22]]} {w w}_{22} / / {w w}_{11})) \end{matrix} - - - - - - ((55))$

上式中W是径向窗函数，V是角度函数，V满足 $t Σ_{- \infty}^{0} V^{2} (t - l) = 1, t &Element; (- \frac{1}{2}, \frac{1}{2}),$ Φ为一维低通窗口的内积：In the above formula, W is the radial window function, V is the angle function, and V satisfies $t Σ_{- \infty}^{0} V^{2} (t - l) = 1, t &Element; (- \frac{1}{2}, \frac{1}{2}),$ Φ is the inner product of a one-dimensional low-pass window:

Φ_j(w₁，w₂)＝φ(2^-jw₁)φ(2^-jw₂)(6)Φ _j (w ₁ , w ₂ )=φ(2 ^-j w ₁ )φ(2 ^-j w ₂ )(6)

上式中φ是基函数，引入相同间隔斜率tanθ_l＝l×2^[-j/2]，l＝-2^[-j/2]，…，2^[-j/2]-1，则In the above formula, φ is the basis function, introduce the same interval slope tanθ _l = l×2 ^[-j/2] , l = -2 ^[-j/2] , ..., 2 ^[-j/2] -1, then

${\overset{~ ~}{U u}}_{j j,, l l} ((ω ω)) = = {\overset{~ ~}{W W}}_{j j} ((ω ω)) {V V}_{j j} (({S S}_{{θ θ}_{11}} ω ω)) - - - - - - ((77))$

其中，剪切矩阵 $S_{θ_{1}} = [\begin{matrix} 1 & 0 \\ - \tan θ & 1 \end{matrix}],$ 离散Curvelet定义为where the shear matrix $S_{θ_{1}} = [\begin{matrix} 1 & 0 \\ - the tan θ & 1 \end{matrix}],$ A discrete Curvelet is defined as

${\overset{~ ~}{φ φ}}_{j j,, l l,, k k} ((x x)) = = 22^{33_{J J} / / 44} {\overset{~ ~}{φ φ}}_{J J} [[{S S}_{{θ θ}_{k k}}^{T T} ((x x - - S S {θ θ}_{l l}^{- - T T} b b))]] - - - - - - ((88))$

b取2离散值(k₁×2^-j，k₂×2^-j/2)。b takes 2 discrete values (k ₁ ×2 ^-j , k ₂ ×2 ^-j/2 ).

离散第二代Curvelet变换定义为The discrete second-generation Curvelet transform is defined as

$c c ((j j,, l l,, k k)) = = &Integral; &Integral; \overset{^^}{f f} ((ω ω)) {\overset{~ ~}{U u}}_{j j} (({S S}_{{θ θ}_{l l}}^{- - 11} ω ω)) {e e}^{i i (({S S}_{{θ θ}_{l l}}^{- - T T} b b,, ω ω))} dω dω - - - - - - ((99))$

由于剪切块 $S_{θ_{l}}^{- T} (k_{1} \times 2^{- j}, k_{2} \times 2^{- j / 2})$ 不是标准矩形，不能采用FFT算法，将上式重写为due to cutout $S_{θ_{l}}^{- T} (k_{1} \times 2^{- j}, k_{2} \times 2^{- j / 2})$ is not a standard rectangle, the FFT algorithm cannot be used, and the above formula can be rewritten as

$c c ((j j,, l l,, k k)) = = &Integral; &Integral; \overset{^^}{f f} ((ω ω)) {\overset{~ ~}{U u}}_{j j} (({S S}_{{θ θ}_{l l}}^{- - 11} ω ω)) {e e}^{i i (({S S}_{{θ θ}_{l l}}^{- - T T} b b,, ω ω))} dω dω = = &Integral; &Integral; \overset{^^}{f f} (({S S}_{{θ θ}_{l l}} ω ω)) {\overset{~ ~}{U u}}_{j j} ((ω ω)) {e e}^{i i ((b b,, ω ω))} dω dω - - - - ((1010))$

最后利用二维离散傅立叶变换Finally, using the two-dimensional discrete Fourier transform

$\overset{^^}{f f} [[{n no}_{11},, {n no}_{22}]] = = {Σ Σ}_{{t t}_{11},, {t t}_{22} = = 00}^{n no - - 11} f f [[{t t}_{11},, {t t}_{22}]] {e e}^{- - i i 22 π π (({n no}_{11} {t t}_{11} + + {n no}_{22} {t t}_{22})) / / n no},, - - n no / / 22 \leq \leq {n no}_{11},, {n no}_{22} < < n no / / 22 - - - - - - ((1111))$

得到运用FFT实现的FDCTGet the FDCT implemented using FFT

${c c}^{D D.} ((j j,, l l,, k k)) = = \underset{{n no}_{11},, {n no}_{22} &Element; &Element; {P P}_{j j}}{Σ Σ} \overset{^^}{f f} [[{n no}_{11},, {n no}_{22} - - {n no}_{11} tan the tan {θ θ}_{l l}]] {\overset{~ ~}{U u}}_{j j} [[{n no}_{11},, {n no}_{22}]] {e e}^{i i 22 π π (({k k}_{11} {n no}_{11} / / {L L}_{11,, l l} + + {k k}_{22} {n no}_{22} / / {L L}_{22,, j j}))} - - - - - - ((1212))$

式中L_1，j和L_2，j分别为窗口的长度和宽度。where L _{1, j} and L _{2, j} are the length and width of the window, respectively.

在公式(12)中，c^D(j，l，k)表示不同分解尺度、不同方向下所对应的Curvelet变换系数，当l＝0且k＝0时，对应的系数表示低频分量，其余系数代表高频分量。In formula (12), c ^D (j, l, k) represents the corresponding Curvelet transform coefficients in different decomposition scales and directions. When l=0 and k=0, the corresponding coefficients represent low-frequency components, and the remaining coefficients represent high frequency components.

步骤二：对非抽样小波变换分解得到的Wavelet系数进行处理，保留多尺度下分解得到的高频成分，将低频成分置零。Step 2: Process the Wavelet coefficients decomposed by the non-sampled wavelet transform, retain the high-frequency components decomposed under multi-scale, and set the low-frequency components to zero.

步骤三：对快速离散Curvelet分解得到的Curvelet系数进行处理，保留多尺度下分解得到的高频成分，将低频成分置零。Step 3: Process the Curvelet coefficients obtained by the fast discrete Curvelet decomposition, retain the high-frequency components obtained by multi-scale decomposition, and set the low-frequency components to zero.

步骤四：分别通过保留的Wavelet系数和Curvelet系数对分解的原始红外图像进行图像重构。Step 4: Perform image reconstruction on the decomposed original infrared image through the retained Wavelet coefficients and Curvelet coefficients respectively.

首先，对Wavelet系数筛选采用的是将低频系数置零的方法，对筛选后得到的Wavelet系数进行非抽样小波逆变换，得到重构的图像。该步骤的处理结果是经过Wavelet变换背景抑制后的重构图像。First, the wavelet coefficients are screened using the method of zeroing the low-frequency coefficients, and the wavelet coefficients obtained after screening are subjected to non-sampling wavelet inverse transformation to obtain the reconstructed image. The processing result of this step is the reconstructed image after background suppression by Wavelet transformation.

其次，对Curvelet系数筛选采用的是将低频系数置零的方法，对筛选后得到的Curvelet系数进行UWT逆变换，得到重构的图像。该步骤的处理过程的处理结果是经过Curvelet变换背景抑制后的重构图像。Secondly, the screening of Curvelet coefficients adopts the method of zeroing the low-frequency coefficients, and performs UWT inverse transformation on the Curvelet coefficients obtained after screening to obtain the reconstructed image. The processing result of the processing in this step is the reconstructed image after background suppression by Curvelet transformation.

步骤五：通过多分辨率分解框架，对步骤四所得到的Wavelet系数重构图像和Curvelet系数重构图像分别进行图像融合，得到最终的红外图像背景抑制结果。Step 5: Through the multi-resolution decomposition framework, image fusion is performed on the Wavelet coefficient reconstructed image and the Curvelet coefficient reconstructed image obtained in Step 4, respectively, to obtain the final infrared image background suppression result.

步骤五中所述的图像融合采用的是基于Wavelet的多分辨率特征融合，具体过程为：The image fusion described in step five adopts Wavelet-based multi-resolution feature fusion, and the specific process is as follows:

首先，利用非抽样小波变换对步骤四所得到的两幅重构图像分别进行分解，得到两组Wavelet变换系数。Firstly, the two reconstructed images obtained in step 4 are decomposed respectively by using non-decimated wavelet transform to obtain two sets of Wavelet transform coefficients.

其次，采用分块的方式，分别在两组Wavelet变换系数上计算对应块的局部区域图像熵。Secondly, the local area image entropy of the corresponding block is calculated on the two sets of Wavelet transform coefficients in a block-by-block manner.

然后，根据局部区域图像熵最大准则，保留两组Wavelet系数对应块的图像熵较大的系数块。当所有分块都利用局部区域图像熵最大准则进行了选择后，利用保留的图像熵较大的分块系数构成完整的图像重构所需的系数。Then, according to the maximization criterion of the image entropy in the local area, the coefficient blocks with larger image entropy of the blocks corresponding to the two groups of Wavelet coefficients are retained. After all the blocks are selected by using the maximization criterion of the image entropy in the local area, the coefficients required for the complete image reconstruction are formed by using the retained block coefficients with larger image entropy.

最后，对融合后的Wavelet系数利用非抽样小波逆变换进行图像重构，得到最终输出的结果图像。在输出的结果图像中，红外背景得到有效地抑制，目标信息(点特征、线特征)得到有效地保存。Finally, image reconstruction is performed on the fused Wavelet coefficients using non-decimated wavelet inverse transform, and the final output image is obtained. In the output result image, the infrared background is effectively suppressed, and the target information (point feature, line feature) is effectively preserved.

上述过程中的特征融合准则采用的是局部区域图像熵最大准则。The feature fusion criterion in the above process adopts the maximum criterion of local area image entropy.

Claims

1, based on the infrared image background suppression method of Wavelet and Curvelet transformation, it is characterized in that it comprises the following steps:

Step 1, performing Wavelet transform decomposition and Curvelet transform decomposition on the original infrared image respectively;

Step 2. For the Wavelet coefficients decomposed in step 1, retain the high-frequency decomposition coefficients at different resolutions, and set the low-frequency decomposition coefficients to zero;

Step 3. For the Curvelet coefficients decomposed in step 1, retain the high-frequency decomposition coefficients at different resolutions, and set the low-frequency decomposition coefficients to zero;

Step 4, performing image reconstruction on the decomposed original infrared image through the retained Wavelet coefficients and Curvelet coefficients respectively;

Step 5. Through the multi-resolution decomposition framework, image fusion is performed on the Wavelet coefficient reconstructed image and the Curvelet coefficient reconstructed image obtained in Step 4, respectively, to obtain the final infrared image background suppression result.

2. The infrared image background suppression method based on Wavelet and Curvelet transform according to claim 1, characterized in that the Wavelet transform decomposition described in step 1 adopts non-sampled Wavelet transform.

3. The infrared image background suppression method based on Wavelet and Curvelet transform according to claim 1, characterized in that the Curvelet transform decomposition described in step 1 adopts the second generation Curvelet transform.

4. The infrared image background suppression method based on Wavelet and Curvelet transformation according to claim 1, characterized in that the image fusion described in step five adopts multi-resolution feature fusion based on Wavelet.

5. The infrared image background suppression method based on Wavelet and Curvelet transform according to claim 1 or 4, characterized in that the feature fusion criterion described in step 5 adopts the criterion of maximizing local area image entropy.