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

Abstract

An infrared image background suppression method which is based on Wavelet and Curvelet transforms relates to an infrared image processing method, and the method is used for solving the problem that the method for suppressing background interference in the infrared image processing field also suppresses the useful target information while realizing the background clutter suppression. The Wavelet transform decomposition and the Curvelet transform decomposition are respectively carried out on an original infrared image; high-frequency decomposition coefficients under different resolutions are respectively retained for the Wavelet coefficient and the Curvelet coefficient, the low-frequency decomposition coefficient is set as zero; image reconstruction is carried out on the decomposed original infrared image by the retained Wavelet coefficient and the Curvelet coefficient respectively; image fusion is respectively carried out on a Wavelet coefficient reconstruction image and a Curvelet coefficient reconstruction image which are obtained by step 4 by a multi-resolution decomposition frame, thereby obtaining the final infrared image background suppressing result.

Description

Infrared image background inhibition method based on Wavelet and Curvelet conversion

Technical field

The present invention relates to a kind of Infrared Image Processing Method, belong to image processing field.

Background technology

An infrared automatic target recognition system generally is made up of functional modules such as target detection, Target Recognition, target acquistion, target with high precision tracking and point of attack selections.Target detection is a critical problem in the precise guidance as foremost processing links in the infrared imaging guidance system.In order to find target as early as possible, make guidance system that the enough reaction time be arranged and improve the early warning distance of the weapons of system, require on far distance, just can detect target, thereby can find target as early as possible.When detection range and imaging viewing field increase,, in imaging plane, also only show as several pixels, even, be called little target less than a pixel even target itself is very big.At this moment, detectable signal relatively a little less than, particularly under the varying background of non-stationary is disturbed, target even flooded by the noise of large amount of complex (clutter), signal noise ratio (snr) of image is extremely low, makes Point Target Detection work become very difficult.Therefore, for outstanding little target, improve signal to noise ratio (S/N ratio), thereby improve target detection probability, the background before the infrared small target image is detected suppresses and noise filtering is very necessary.In the automatic target recognition system, background suppresses and noise-cut is referred to as the preceding filter preprocessing of detection.The final purpose that infrared image background suppresses is in order to eliminate noise jamming, to preserve target information.

The method that in the infrared image processing field background interference is suppressed at present roughly can be divided into two big classes: the Space Time territory is handled and frequency domain/transform domain is handled.The former typical method comprises linear background forecast, Top-hat conversion etc., and they realize the prediction of infrared image background by airspace filter, utilizes prognostic chart and original graph to do difference and obtains the result that background suppresses; The latter's typical method comprises high-pass filtering, Wavelet conversion etc., mainly be by decomposing to the analysis of infrared image frequency content or by multiresolution, removal or inhibition comprise the frequency component or the conversion coefficient of background information, thereby reach the purpose that suppresses background.In said method, the Wavelet conversion shows preferable performance with multiple dimensioned decomposition of its signal that has and descriptive power in infrared image background suppresses.But the Wavelet conversion only is suitable for describing the signal with zero dimension singularity, that is to say that the Wavelet conversion also can curb useful target information when realizing that background clutter suppresses, particularly the line feature (as road, bridge, airport, harbour etc. the zone the edge).

Summary of the invention

The present invention provides a kind of infrared image background inhibition method based on Wavelet and Curvelet conversion for solving the problem that also can curb useful target information when realizing the background clutter inhibition of the method existence that in the infrared image processing field background interference is suppressed.The present invention includes following steps:

Step 1, original infrared image carried out respectively the Wavelet conversion is decomposed and the Curvelet conversion is decomposed;

Step 2, to decomposing the Wavelet coefficient obtain in the step 1, keep the high frequency coefficient of dissociation under the different resolution, with the zero setting of low frequency coefficient of dissociation;

Step 3, to decomposing the Curvelet coefficient obtain in the step 1, keep the high frequency coefficient of dissociation under the different resolution, with the zero setting of low frequency coefficient of dissociation;

Step 4, by Wavelet coefficient and the Curvelet coefficient that keeps the original infrared image that decomposes is carried out image reconstruction respectively;

Step 5, decompose framework, step 4 resulting Wavelet coefficient reconstructed image and Curvelet coefficient reconstructed image are carried out image co-registration respectively, obtain final infrared image background and suppress the result by multiresolution.

Principle of work of the present invention is: utilize Wavelet and Curvelet to carry out the background inhibition to importing original infrared image respectively, suppress among the result in background, can be when suppressing background in the Wavelet result image, preserve the impact point characteristic information preferably, can preserve the score characteristic information preferably in the Curvelet result image.To the image after Wavelet and the inhibition of Curvelet background, carry out the Feature Fusion based on Wavelet multiresolution analysis and local area image entropy criterion, the background that is effectively suppressed has been preserved the result images of impact point characteristic information and line characteristic information simultaneously preferably.In order to give full play to Wavelet and Curvelet signal description ability, adopted non-sampling Wavelet conversion (Undecimated Wavelet Transform among the present invention, UWT) and second generation Curvelet conversion (fast discrete Curvelet conversion, Fast Discrete Curvelet Transform, FDCT).

The present invention compared with prior art has following advantage:

(1) broken through the existing deficiency that exists based on the infrared image background inhibition method of Wavelet conversion, promptly these class methods only can suppress background preferably, but have also lost target information simultaneously.And the inventive method can suppress background effectively, can preserve target information well again, has so just reached the real purpose that infrared image background suppresses.

(2) aspect the target information preservation, the inventive method can be on the basis of preserving target information effectively, further guarantee the diversity of target information, promptly can either preserve the some characteristic information of target effectively, can preserve the line characteristic information of target again effectively.The preservation of some characteristic information and line characteristic information is the technological difficulties that infrared image background suppresses, and also is the key of follow-up infrared image target detection and identification.

(3) owing to utilized Wavelet and the Curvelet multiple dimensioned description characteristic to image simultaneously, it is good to suppress background effect, can improve target detection ability and bearing accuracy as systems such as infrared target detection and identification, scene of a fire infrared image monitorings greatly.

Description of drawings

Fig. 1 is a Flame Image Process process flow diagram of the present invention; Fig. 2 is a workflow diagram of the present invention; The contrast that Fig. 3 decomposes for non-sampling Wavelet of the present invention conversion and tradition sampling Wavelet transfer pair one-dimensional signal: the comparison diagram that tradition sampling Wavelet conversion and non-sampling Wavelet transfer pair one-dimensional signal carry out three layers of decomposition, wherein (a) is traditional DWT decomposing schematic representation, (b) is non-sampling Wavelet (UWT) decomposing schematic representation; Fig. 4 is the frequency space area dividing figure of fast discrete Curvelet of the present invention conversion.

Embodiment

Embodiment one: referring to Fig. 1～Fig. 4, present embodiment is made up of following steps:

Step 1: original infrared image is carried out Wavelet conversion decomposition and Curvelet decomposition respectively.

At first, the original infrared image to input carries out non-sampling Wavelet conversion (UndecimatedWavelet Transform, UWT) decomposition.

What traditional wavelet transform adopted when decomposed signal is the sampling operation, and this makes the part temporal signatures of original signal not to be retained in the decomposition result, and decomposition result is that translation is variable.In order to overcome this defective and to obtain the characteristic of more complete analyzed signal, the present invention has adopted non-sampling wavelet transformation, its ultimate principle is signal not to be sampled, so it is translation invariant, and abundant temporal signatures information and accurate frequency localization information is provided.

(Undecimated Wavelet Transform is that a kind of nothing extracts discrete small wave converting method UWT) to non-sampling wavelet transformation, and it does not carry out the down-sampling operation in conversion, have isotropy and translation invariance.Concrete operations are to utilize high and low pass filter (h, g) respectively image is carried out Filtering Processing, each grade processing just obtains the low-and high-frequency coefficient after the conversion afterwards, then to bank of filters (h, g) carry out interlacing respectively every listing sampling, the concrete method of sampling is when each grade decomposition, and the upper level wave filter is carried out interlacing, inserts 0 every row and handle, and makes one times of its frequency filtering expanded range.Image by the UWT conversion just can obtain a series of high fdrequency component and low frequency component like this, i.e. S=(S ¹, S ²..., S ^J, A ^J), wherein, S ^JBe the high frequency coefficient that decomposes the j layer, A _JIt is the low frequency coefficient that decomposes the J layer.Its reconstructing method is as the formula (1):

$R=\underset{j=0}{\overset{J}{\mathrm{\Σ}}}\frac{1}{2}[(h^j*A_{j+1})+(g^j*S_{j+1})]---\left(1\right)$

Wherein (h g) is respectively the reconstruct high-pass and low-pass filter, and in order to realize complete image reconstruction, it must satisfy formula (2) relation with resolution filter in frequency domain:

$H\left(z^{-1}\right)\tilde{H}\left(z\right)+G\left(z^{-1}\right)\tilde{G}\left(z\right)=1---\left(2\right)$

Wherein, H, G are respectively the frequency coefficients of resolution filter,

It is respectively the frequency coefficient of reconfigurable filter.

Secondly, the original infrared image to input carries out fast discrete Curvelet decomposition.

Image is being carried out Curvelet when decomposing, the present invention adopts be fast discrete Curvelet conversion (Fast Discrete Curvelet Transform, FDCT), i.e. two generation Curvelet conversion.FDCT is the two generation Curvelet conversion that grow up on original Curvelet conversion basis, simplifies on its principle, calculates quicker.Generation Curvelet conversion is that the different scale sub-band images adopted different big or small piecemeals after image was carried out sub-band division, each piece is carried out the Ridgelet decomposition obtain the Curvelet coefficient, 7 parameters are arranged in its transition structure, the mathematical analysis more complicated that causes conversion, and the cascading windows that adopt have realized strengthening the difficulty that conversion realizes.FDCT reduces to 3 with parameter, adopts simpler, more transparent structure to construct the Curvelet conversion, need not to use Ridgelet in the implementation procedure and decomposes, and adopt fast fourier transform to quicken, and has simplified the Curvelet conversion process greatly.

On the basis of Curvelet conversion, FDCT can followingly explain:

In the formula

Discrete Curvelet (subscript D represents to disperse), formula (3) is to copy the continuous Curvelet variation of formula.

FDCT adopts concentric square zone

Replace, to be fit to image processing under two-dimentional cartesian coordinate system when discretize is handled, its partitioned mode such as Fig. 4 show.

Local window under the cartesian coordinate system of FDCT definition is:

${\tilde{U}}_j\left(w\right)={\tilde{W}}_j\left(w\right)V_j\left(w\right)---\left(4\right)$

Wherein

$\left\{\begin{array}{c}{\tilde{W}}_j\left(w\right)=\sqrt{{\mathrm{\Φ}}_{j+1}^2\left(w\right)-{\mathrm{\Φ}}_j^2\left(w\right)},j\≥1\\ V_j\left(w\right)=V(2^{[-j/2]}{w}_2/w_1)\end{array}\right.---\left(5\right)$

W is a window function radially in the following formula, and V is an angle function, and V satisfies $t\underset{-\∞}{\overset{0}{\mathrm{\Σ}}}{V}^2(t-l)=1,t\∈(-\frac{1}{2},\frac{1}{2}),$ Φ is the inner product of one dimension low pass window:

Φ _j(w ₁，w ₂)＝φ(2 ^-jw ₁)φ(2 ^-jw ₂)(6)

φ is a basis function in the following formula, introduces same intervals slope tan θ _l=l * 2 ^[j/2], l=-2 ^[j/2]..., 2 ^[j/2]-1, then

${\tilde{U}}_{j,l}\left(\mathrm{\ω}\right)={\tilde{W}}_j\left(\mathrm{\ω}\right)V_j\left(S_{{\mathrm{\θ}}_1}\mathrm{\ω}\right)---\left(7\right)$

Wherein, shear matrix $S_{{\mathrm{\θ}}_1}=\left[\begin{array}{cc}1& 0\\ -\tan \mathrm{\θ}& 1\end{array}\right],$ Discrete Curvelet is defined as

${\widetilde{\mathrm{\φ}}}_{j,l,k}\left(x\right)=2^{3_J/4}{\widetilde{\mathrm{\φ}}}_J\left[S_{{\mathrm{\θ}}_k}^T(x-S{\mathrm{\θ}}_l^{-T}b)\right]---\left(8\right)$

B gets 2 discrete value (k ₁* 2 ^-j, k ₂* 2 ^-j/2).

Discrete second generation Curvelet transform definition is

$c(j,l,k)=\∫\hat{f}\left(\mathrm{\ω}\right){\tilde{U}}_j\left(S_{{\mathrm{\θ}}_l}^{-1}\mathrm{\ω}\right)e^{i(S_{{\mathrm{\θ}}_l}^{-T}b,\mathrm{\ω})}\mathrm{d\ω}---\left(9\right)$

Because cutout $S_{{\mathrm{\&theta;}}_l}^{-T}(k_1\&times;2^{-j},{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">k</figure-callout>}}_2\&times;2^{-j/2})$ Not standard rectangular, can not adopt fft algorithm, following formula is rewritten as

$c(j,l,k)=\&Integral;\hat{f}\left(\mathrm{\&omega;}\right){\tilde{U}}_j\left(S_{{\mathrm{\&theta;}}_l}^{-1}\mathrm{\&omega;}\right)e^{i(S_{{\mathrm{\&theta;}}_l}^{-T}b,\mathrm{\&omega;})}\mathrm{d\&omega;}=\&Integral;\hat{f}\left(S_{{\mathrm{\&theta;}}_l}\mathrm{\&omega;}\right){\tilde{U}}_j\left(\mathrm{\&omega;}\right)e^{i(b,\mathrm{\&omega;})}\mathrm{d\&omega;}--\left(10\right)$

Utilize two dimension discrete fourier transform at last

$\hat{f}[n_1,n_2]=\underset{t_1,{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">t</figure-callout>}}_2=0}{\overset{n-1}{\mathrm{\&Sigma;}}}f[t_1,t_2]e^{-\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">i</figure-callout>}2\mathrm{\&pi;}(n_1{t}_1+n_2{t}_2)/n},-n/2\&le;n_1,{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">n</figure-callout>}}_2<n/2---\left(11\right)$

The FDCT that the FFT that finds application realizes

$c^D(j,l,k)=\underset{n_1,{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">n</figure-callout>}}_2\&Element;P_j}{\mathrm{\&Sigma;}}\hat{f}[n_1,n_2-n_1\tan {\mathrm{\&theta;}}_l]{\tilde{U}}_j[n_1,n_2]e^{i2\mathrm{\&pi;}(k_1{n}_1/L_{1,l}+k_2{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">n</figure-callout>}}_2/{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="A20081013753300111.png"\; state="\{\{state\}\}">L</figure-callout>}}_{2,j})}---\left(12\right)$

L in the formula _{1, j}And L _{2, j}Be respectively the length and the width of window.

In formula (12), c ^D(j, l, k) pairing Curvelet conversion coefficient under expression different decomposition yardstick, the different directions, when l=0 and k=0, corresponding coefficient is represented low frequency component, all the other coefficients are represented high fdrequency component.

Step 2: non-sampling wavelet transformation is decomposed the Wavelet coefficient that obtains handle, keep the radio-frequency component that multiple dimensioned decomposition down obtains, with low-frequency component zero setting.

Step 3: fast discrete Curvelet is decomposed the Curvelet coefficient that obtains handle, keep the radio-frequency component that multiple dimensioned decomposition down obtains, with low-frequency component zero setting.

Step 4: by Wavelet coefficient and the Curvelet coefficient that keeps the original infrared image that decomposes is carried out image reconstruction respectively.

At first, what Wavelet coefficient screening was adopted is method with low frequency coefficient zero setting, and the Wavelet coefficient that obtains after the screening is carried out non-sampling wavelet inverse transformation, obtains the image of reconstruct.The result of this step is the reconstructed image after suppressing through the Wavelet changing background.

Secondly, screening is adopted to the Curvelet coefficient is method with low frequency coefficient zero setting, and the Curvelet coefficient that obtains after the screening is carried out the UWT inverse transformation, obtains the image of reconstruct.The result of the processing procedure of this step is the reconstructed image after suppressing through the Curvelet changing background.

Step 5: decompose framework by multiresolution, step 4 resulting Wavelet coefficient reconstructed image and Curvelet coefficient reconstructed image are carried out image co-registration respectively, obtain final infrared image background and suppress the result.

The multiresolution features that is based on Wavelet that image co-registration described in the step 5 adopts merges, and detailed process is:

At first, utilize non-sampling wavelet transformation that resulting two width of cloth reconstructed images of step 4 are decomposed respectively, obtain two groups of Wavelet conversion coefficients.

Secondly, adopt the mode of piecemeal, on two groups of Wavelet conversion coefficients, calculate the regional area image entropy of corresponding blocks respectively.

Then, according to regional area image entropy maximal criterion, keep the bigger coefficient block of image entropy of two groups of Wavelet coefficient corresponding blocks.After all piecemeals all utilize regional area image entropy maximal criterion to carry out selecting, utilize the bigger blocking factor of image entropy that keeps to constitute the complete required coefficient of image reconstruction.

At last, utilize non-sampling wavelet inverse transformation to carry out image reconstruction, the result images of finally being exported to the Wavelet coefficient after merging.In the result images of output, infrared background is suppressed effectively, and target information (some feature, line feature) is preserved effectively.

What the Feature Fusion criterion in the said process adopted is regional area image entropy maximal criterion.

Claims (5)

1,, it is characterized in that it may further comprise the steps based on the infrared image background inhibition method of Wavelet and Curvelet conversion:

Step 1, original infrared image carried out respectively the Wavelet conversion is decomposed and the Curvelet conversion is decomposed;

Step 2, to decomposing the Wavelet coefficient obtain in the step 1, keep the high frequency coefficient of dissociation under the different resolution, with the zero setting of low frequency coefficient of dissociation;

Step 3, to decomposing the Curvelet coefficient obtain in the step 1, keep the high frequency coefficient of dissociation under the different resolution, with the zero setting of low frequency coefficient of dissociation;

Step 4, by Wavelet coefficient and the Curvelet coefficient that keeps the original infrared image that decomposes is carried out image reconstruction respectively;

Step 5, decompose framework, step 4 resulting Wavelet coefficient reconstructed image and Curvelet coefficient reconstructed image are carried out image co-registration respectively, obtain final infrared image background and suppress the result by multiresolution.

2, the infrared image background inhibition method based on Wavelet and Curvelet conversion according to claim 1, what it is characterized in that the Wavelet conversion decomposition employing described in the step 1 is non-sampling Wavelet conversion.

3, the infrared image background inhibition method based on Wavelet and Curvelet conversion according to claim 1, what it is characterized in that the Curvelet conversion decomposition employing described in the step 1 is second generation Curvelet conversion.

4, the infrared image background inhibition method based on Wavelet and Curvelet conversion according to claim 1 is characterized in that the multiresolution features that is based on Wavelet that the image co-registration described in the step 5 adopts merges.

5, according to claim 1 or 4 described infrared image background inhibition methods based on Wavelet and Curvelet conversion, what it is characterized in that the Feature Fusion criterion employing described in the step 5 is regional area image entropy maximal criterion.

Images