CN104517267A - Infrared image enhancement and reestablishment method based on spectra inversion - Google Patents

Abstract

The invention discloses an infrared image enhancement and reestablishment method based on spectra inversion, belongs to the field of infrared image processing, and particularly relates to a new infrared image enhancement and detail reestablishment method and a new computational imaging method. The method uses a seismic trace high-resolution reestablishment thought in seismic signal processing proposed recently for reference to carry out the spectra inversion (reflected energy reconstruction) of infrared images. The method carries out the spectra inversion in dominant frequency bands by establishing an infrared reflected energy model and reflected energy odd-even decomposition model, and finally realizes the improvement of infrared image detail resolution ratio. The method can highlight the reflected energy details of weak and small targets on the basis of without increasing the image space resolution, thus improving the target contrast ratio and the energy feature. The method is beneficial for subsequent detection and tracking of the infrared weak and small targets, and has a wider application prospect.

Description

A kind of infrared image enhancement based on " spectrum " inverting and method for reconstructing

Technical field

The invention belongs to infrared image processing field, particularly a kind of new infrared image enhancement and details method for reconstructing and new calculating formation method.The method uses for reference seismic trace super-resolution reconstruction thought in the seismic data processing proposed in the recent period, carries out " spectrum " inverting (reflected energy reconstruct) of infrared image.

Background technology

Infrared imagery technique is widely used in the field such as Modern Traffic, safety monitoring, and in infrared detection system, infrared target detection and tracking are its gordian technique and nucleus module.Because remote infrared imaging often brings, picture contrast is low, when signal to noise ratio is low, imaging area is little for noise, the defect such as amorphism and textural characteristics, need to carry out pre-service to the infrared image collected, improve the detail resolution of image, to suppress background energy, to improve target area energy and outstanding target detail, be conducive to the detection and tracking of follow-up infrared small object.

Traditional super-resolution image reconstruction (Super resolution image re-construction, SRIR/SR) method with signal transacting and image procossing is referred to, by the mode of software algorithm by existing low resolution (Low-resolution, LR) image converts high resolving power (High-resolution to, HR) technology of image, main method is divided into the super-resolution rebuilding based on Image Reconstruction, and based on the super-resolution rebuilding etc. learnt.This technology by utilizing one group of inferior quality, low-resolution image (or motion sequence) produces single width high-quality, high-definition picture.Classic method often brings the change of image space size, and can not strengthen the energy feature of area-of-interest targetedly.

In seismic prospecting signal transacting, spectrum inverting is the inversion method of a kind of meticulous identification thin layer and reflection coefficient.Spectrum inverting is based upon on the basis of seismic signal spectral factorization, and by spectral factorization, we extract seismic trace instantaneous attribute, and these attributes can be used for suitably improving the resolution of seismic data.In frequency field, carry out estimation and the reconstruct of reflection coefficient, to improve the resolution of inversion result further, be spectrum inverting.

1980s, the concept of spectral factorization is suggested.The complex trace analysis that the people such as Taner (1979) propose defines theory of spectrum dissolution basis the earliest.Along with uncertainty principle is introduced into digital processing field, Spectral Decomposition Technique starts to develop rapidly.Use for reference the method with different time-frequency basis function analytic signals in digital signal processing, the methods such as Gabor transformation, continuous wavelet transform, wavelet transform, S-transformation are introduced seismic data analysis by physical prospecting scholars in succession.Applying along with these new algorithms, the extraction result of seismic data instantaneous attribute there has also been significant improvement.Afterwards, Portniaguine (2004) proposes seismic signal inverse spectral decomposition method, the people such as Sinha, Castagna (2005) propose method continuous wavelet transform being used for earthquake spectral factorization etc., and the proposition of these methods facilitates the development of seismic data Spectral Decomposition Technique.

Spectrum inverting is the combination of spectral factorization and inversion algorithm.Through years of researches and development, spectrum inversion method is proved to be as a kind of effective, method that can improve seismic data image thin layer resolution.There is no noise and known seismic wavelet ideally, the method can identify that thickness is less than the thin layer of tuning thickness in theory, and accurately can depict the border on stratum.The objective function of spectrum inverting has good convergence and restriction ability, by adjusting position and the size of reflection coefficient, under the constraint of frequency domain objective function, can obtain high resolving power reflection coefficient sequence.

Although spectrum inversion method is from proposing to experienced by till now the development in more than ten years, in gordian technique and application thereof, also having many problems to need to go further investigation, it or a kind of new method and technology can be said.Look back the developing history of spectrum inverting, can following several stages be divided into:

Partyka (1999) etc. find that spectral factorization method is applied in seismic data interpretation, suitably can improve the resolution of seismic data; 2004, Spectral Decomposition Technique in conjunction with seismic inversion, was proposed spectrum inversion theory by Partyka etc.Satinder Chopra, JohnP.Castagna etc. (2006) the local spectrum data that spectral factorization obtains carries out inverting, calculate thin bed reflections coefficient, the resolution of inversion result is better than conventional inverting, and indicates that the method has without the need to advantages such as prior model, reflection coefficient hypothesis, layer restrain, Log-constraineds.Charles I.Puryear and John P.Castagna (2008) resolves into even component and odd component reflection coefficient, derive the fundamental formular of spectrum inverting, and demonstrate the method by modeling computation and can be used to differentiate underground and be less than the thin layer of tuning thickness; Sanyi Yuan etc. (2009) discuss present stage spectrum inverting Problems existing and applicable elements thereof theoretically, and have carried out detailed analysis to the ill-posedness of spectrum inverting.Spectrum inversion method is applied to real data by Satinder.Chopra and John P.Castagna etc. (2009), and calculated sparse reflection coefficient, this not only demonstrates the feasibility of spectrum inversion method, also points out application prospect and the future thrust of composing inverting.

Summary of the invention

The present invention designs and a kind ofly uses for reference the image processing method of seismic prospecting signal spectrum inversion theory based on wavelet decomposition.By setting up infrared image reflected energy model, adopt " spectrum " inversion method, when not changing spatial resolution, the detail resolution realizing image strengthens.The method is the identification of the subsequent treatment of Infrared Image Information, such as Weak target, and the classification of background etc. provide more useful and efficient information, can improve infrared target detection and discrimination.

A kind of infrared image enhancement based on " spectrum " inverting of the present invention and method for reconstructing, the method comprises:

Step 1: obtain a pending infrared image;

Step 2: adopt small echo to build infrared image and decompose dictionary, the atom in dictionary is sparse base or orthogonal basis;

Step 3: imagery exploitation dictionary step 1 obtained decomposes, and obtains some subbands, and obtains the coefficient of each subband;

Step 4: set up each sub belt energy model;

Step 5: carry out the decomposition of odd even component to each sub-band coefficients, obtains odd component coefficient and even component coefficient, and can construct the energy model of odd even component subband and correspondence thereof respectively;

Step 6: set up infrared image detail resolution evaluation criterion;

Step 7: odd component coefficient and the even component coefficient of each subband step 5 obtained are multiplied by different transformation factors, change its size, then utilize the reconstruct of the coefficient after changing infrared image, after the evaluation criterion set up by step 6 judges reduction, whether image resolution ratio reaches requirement;

Step 8: to the method that step 7 is identical, details strengthening is carried out to the infrared image collected under several similar situations according to step 1, obtain the component coefficient transformation factor of different images, obtain a component coefficient transformation factor generally used according to the component coefficient transformation factor obtained again, the details for the infrared image obtained under similar situation is strengthened.

Orthogonal wavelet is adopted to build dictionary in wherein said step 2.

Utilize dictionary to be high-frequency sub-band and low frequency sub-band to acquisition picture breakdown in wherein said step 3, and obtain the sub-band coefficients of its correspondence;

Wherein said step 6 sets up infrared image detail resolution evaluation criterion: C=C _def× C _sNR, wherein: C _deffor the sharpness of image, C _sNRfor the signal to noise ratio (S/N ratio) of image;

The concrete steps of wherein said step 7 are:

Step 7-1: set up forward model

$H_1{K}_{1}M{K}_2^H{H_2}^H+n=F$

M is the optimum sub-band coefficients of requirement, K ₁and K ₂for inverse transformation operator, H ₁and H ₂for the degeneracy operator of image, such as, fuzzy, neighbourhood noise caused by camera imaging quality, atmospheric disturbance etc., F represents Image Sub-Band, and n is a small disturbance, represents the otherness of forward model and actual signal;

Step 7-2: obtain objective function according to step 7-1:

$\underset{M}{\min }{\left|\right|H_1{K}_{1}M{K}_2^H{H_2}^H-F\left|\right|}_2+\mathrm{\α}\left|\right|M\left|\right|,s.t.C\≥\mathrm{\δ}$

Wherein δ represents infrared image detail resolution Evaluation threshold, and α is regularization factors.

A kind of infrared image enhancement based on " spectrum " inverting of the present invention and method for reconstructing, belong to infrared image processing field, particularly a kind of new infrared image enhancement and details method for reconstructing and new calculating formation method.The method, by setting up infrared external reflection energy model and the odd, even decomposition model of reflected energy, is carried out the spectrum inverting in dominant frequency band, is finally realized the raising of infrared image detail resolution.This invention can not increase on the basis of image spatial resolution, and " reflection " energy details of outstanding Weak target, improves target contrast and energy feature.Be conducive to the detection and tracking of follow-up infrared small object, there is larger application prospect.

Accompanying drawing explanation

Fig. 1 is the odd even component decomposing schematic representation of the signal containing two pulses;

Fig. 2 is the practical IR image that a width contains Weak target;

Fig. 3 is the result figure that infrared image different frequency bands subgraph carries out the decomposition of odd even component;

Fig. 4 adopts the method proposed in the present invention, the infrared image obtained " spectrum " inversion result figure;

Fig. 5 is the energy profile of infrared image " spectrum " inversion result, and encircled portion represents the infrared small object energy be enhanced;

Fig. 6 is method flow diagram.

Embodiment

Below in conjunction with accompanying drawing, the present invention is specifically described.

Step 1: input a pending infrared image f (x, y), its size is M × N;

Step 2: build infrared image and decompose dictionary, the atom in dictionary is sparse base or orthogonal basis.In the present invention, orthogonal wavelet (as Haar small echo) is adopted to build dictionary;

Step 3: the image read in is carried out the decomposition of j multi-scale wavelet:

In wavelet decomposition, calculate high frequency coefficient by following formula:

$W_{\mathrm{\ψ}}(j,m,n)=\frac{1}{\sqrt{\mathrm{MN}}}\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y){{\mathrm{\ψ}}^i}_{j,m,n}(x,y)---\left(2\right)$

And low frequency coefficient

Step 4: according to the result of step 3, builds infrared image low-and high-frequency reflected energy model (sub belt energy model).Owing to adopting wavelet sub-band energy model here, high-frequency energy is expressed as

${E^j}_{\mathrm{\ψ}}=\frac{1}{M\×N}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\left[W_{\mathrm{\ψ}}(j,m,n)\right]}^2---\left(4\right)$

Here j represents wavelet decomposition progression.

Low frequency energy is expressed as

Here j ₀represent first order wavelet decomposition progression.

Step 5: choose high-frequency sub-band coefficient, carries out the decomposition of odd even component.Here for the signal containing two pulses, the Parity-decomposition principle (see Fig. 1) of signal is described.In the spectrum inverting of seismic signal, we think that the even component of reflection coefficient can improve thin layer resolution, and odd component weakens thin layer resolution.For the infrared image (see Fig. 2) containing Weak target, its high-frequency sub-band coefficient is carried out Parity-decomposition by us, obtains odd component sub-band coefficients and even component sub-band coefficients, and can synthesize odd even component high frequency subgraph (see Fig. 3);

Step 6: set up infrared image detail resolution and strengthen evaluation criterion.For the needs that feature and the succeeding target of infrared image detect, take sharpness and signal to noise ratio as evaluation criterion.The sharpness of image is expressed as

$C_{\mathrm{def}}=\frac{1}{\mathrm{MN}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{\left(\mathrm{\Δ}{f}_x(i,j)\right)}^2+{\left(\mathrm{\Δ}{f}_y(i,j)\right)}^2}---\left(6\right)$

Wherein

△f _x(i,j)＝f(i,j)-f(i-1,j)

(7)

△f _y(i,j)＝f(i,j)-f(i,j-1)

Signal to noise ratio is expressed as

$C_{\mathrm{SNR}}=\frac{|{\mathrm{\μ}}_t-{\mathrm{\μ}}_b|}{{\mathrm{\σ}}_b}---\left(8\right)$

Wherein μ _trepresent the average pixel value of target area, μ _band σ _brepresent mean value and the standard deviation of target surrounding pixel point respectively.Consider above two kinds of evaluation criterions, set up the constraint condition in this problem

$C=\frac{C_{\mathrm{def}}\×C_{\mathrm{SNR}}}{\mathrm{\β}}---\left(9\right)$

Wherein β is a constant, can adjust according to actual needs.

Step 7: derived object function, solves " puppet " sub-band coefficients that can strengthen infrared small object imaging plane energy.Image containing infrared small object can be considered as superposing of the projection of target area on imaging plane and background area, sets up forward model accordingly

$H_1{K}_{1}M{K}_2^N{H_2}^H+n=F---\left(10\right)$

M is amount to be asked, the optimum sub-band coefficients namely required by us, K ₁and K ₂be the inverse transformation operator corresponding to conversion step (3) Suo Shi, H ₁and H ₂represent the degeneracy operator of image, such as, fuzzy, neighbourhood noise caused by camera imaging quality, atmospheric disturbance etc., F represents the subband treating inverting.Solve following objective function:

$\underset{M}{\min }{\left|\right|H_1{K}_{1}M{K}_2^H{H_2}^H-F\left|\right|}_2+\mathrm{\α}\left|\right|M\left|\right|,s.t.C\≥\mathrm{\δ}---\left(11\right)$

Wherein δ represents the degree that infrared image detail resolution is enhanced, and concrete numerical value can be determined according to actual needs.Said process is infrared image " spectrum " refutation process, can improve " puppet " reflected energy on Weak target imaging plane; Inversion method can adopt nonlinear optimization algorithm, as simulated annealing, random hill-climbing algorithm etc., tries to achieve the optimum sub belt energy coefficient that can strengthen infrared small object detail resolution.Utilize " puppet " sub-band coefficients of trying to achieve, the high and low frequency subgraph made new advances can be reconstructed;

Step 8: the result utilizing step 7, carries out Image Reconstruction, and be enhanced the infrared image of detail resolution, Output rusults (see Fig. 4, Fig. 5).In Fig. 5, encircled portion represents the infrared small object energy obtaining enhancing.

Claims (5)

1., based on infrared image enhancement and the method for reconstructing of " spectrum " inverting, the method comprises:

Step 1: obtain a pending infrared image;

Step 2: adopt small echo to build infrared image and decompose dictionary, the atom in dictionary is sparse base or orthogonal basis;

Step 3: imagery exploitation dictionary step 1 obtained decomposes, and obtains some subbands, and obtains the coefficient of each subband;

Step 4: set up each sub belt energy model;

Step 5: carry out the decomposition of odd even component to each sub-band coefficients, obtains odd component coefficient and even component coefficient, and can construct the energy model of odd even component subband and correspondence thereof respectively;

Step 6: set up infrared image detail resolution evaluation criterion;

Step 7: odd component coefficient and the even component coefficient of each subband step 5 obtained are multiplied by different transformation factors, change its size, then utilize the reconstruct of the coefficient after changing infrared image, after the evaluation criterion set up by step 6 judges reduction, whether image resolution ratio reaches requirement;

Step 8: to the method that step 7 is identical, details strengthening is carried out to the infrared image collected under several similar situations according to step 1, obtain the component coefficient transformation factor of different images, obtain a component coefficient transformation factor generally used according to the component coefficient transformation factor obtained again, the details for the infrared image obtained under similar situation is strengthened.

2. a kind of infrared image enhancement based on " spectrum " inverting as claimed in claim 1 and method for reconstructing, is characterized in that adopting orthogonal wavelet to build dictionary in described step 2.

3. a kind of infrared image enhancement based on " spectrum " inverting as claimed in claim 1 and method for reconstructing, is characterized in that utilizing dictionary to be high-frequency sub-band and low frequency sub-band to acquisition picture breakdown in described step 3, and obtains the sub-band coefficients of its correspondence.

4. a kind of infrared image enhancement based on " spectrum " inverting as claimed in claim 1 and method for reconstructing, is characterized in that described step 6 is set up infrared image detail resolution evaluation criterion and is: C=C _def× C _sNR, wherein: C _deffor the sharpness of image, C _sNRfor the signal to noise ratio (S/N ratio) of image.

5. a kind of infrared image enhancement based on " spectrum " inverting as claimed in claim 1 and method for reconstructing, is characterized in that the concrete steps of described step 7 are:

Step 7-1: set up forward model

M is the optimum sub-band coefficients of requirement, K ₁and K ₂for inverse transformation operator, H ₁and H ₂for the degeneracy operator of image, such as, fuzzy, neighbourhood noise caused by camera imaging quality, atmospheric disturbance etc., F represents Image Sub-Band, and n is a small disturbance, represents the otherness of forward model and actual signal;

Step 7-2: obtain objective function according to step 7-1:

Wherein δ represents infrared image detail resolution Evaluation threshold, and α is regularization factors.