CN103886590B - A kind of push-broom type remote sensing camera Atomatic focusing method based on Wavelet Packet Energy Spectrum - Google Patents

Abstract

The invention discloses a kind of push-broom type remote sensing camera Atomatic focusing method based on Wavelet Packet Energy Spectrum, comprise the steps: first greyscale image transitions and brightness normalization pretreatment；Secondly image is carried out four layers of WAVELET PACKET DECOMPOSITION, choose the frequency domain frequency range matrix of wavelet coefficients vertical with camera motion direction；Then each frequency range Wavelet Packet Energy Spectrum exponential quantity and sharpness evaluation function value are calculated；Last camera auto-focusing adjusts, and selecting the focusing position corresponding to definition evaluation of estimate maximum is final camera blur-free imaging position, terminates auto-focusing.The present invention can be by being weighted summation to the result of calculation of remote sensing images motion vertical direction out of focus each frequency range Wavelet Packet Energy Spectrum, realize the focusing evaluation for push-broom type remote sensing camera, evaluation methodology has good accuracy and monotonicity, and accomplishes unrelated with picture material.

Description

A kind of push-broom type remote sensing camera Atomatic focusing method based on Wavelet Packet Energy Spectrum

Technical field

The invention belongs to remotely sensed image technical field, relate to a kind of push-broom type based on Wavelet Packet Energy Spectrum distant Sense automatic focusing method of camera.

Background technology

Along with the development of Aid of Space Remote Sensing Technology, remote sensing images environmental monitoring, resource investigation, mapping, The each side such as military surveillance, play the most important effect.Remote sensing camera is that digital remote sensing image obtains Key, in order to ensure space remote sensing camera in complex environment can blur-free imaging, space remote sensing camera Need to be equipped with a set of focusing system, revise focusing position in time to ensure image quality.

Push-broom type remote sensing camera is the space remote sensing camera that a class is conventional, and this kind of camera is not a width Ground exposure, but continuous exposure, so this kind of camera need not shutter, structure is relatively easy.Along with electricity The fast development of sub-technology, Atomatic focusing method based on image procossing becomes the focusing of push-broom type remote sensing camera The study hotspot of technology, by the focusing definition values being calculated image of evaluation methodology thus instruct phase Machine carries out work of focusing.

Traditional focusing evaluation methodology is required for scenery and keeps static, and space camera is constantly in motion shape State, thus cannot directly these focusing evaluation methodologys be applied in space camera；Meanwhile, push-broom type is distant The scenery feeling camera captured at any time is all different, there is not overlapping region between frame and frame. The evaluation methodology of push-broom type remote sensing camera focusing at present mainly has: based on differential map as autocorrelative focusing is evaluated Method, focusing evaluation methodology based on line spread function and focusing evaluation methodology based on power spectrum.Its In focusing evaluation methodology based on power spectrum be the Autofocus Technology being best suited for push-broom type remote sensing camera, By wavelet transformation, ask for the wavelet coefficient under the different scale vertical with the direction of motion, and to the direction These wavelet coefficients of lower different scale are weighted summation, thus as definition evaluation index, not only Efficiently solve as moving mismatch problems, and all have very at aspects such as robustness, correctness, accuracies Good performance.But when being evaluated the different scenes that fuzzy quantity is identical, its evaluation of estimate still cannot Reach identical value, when especially the scene structure bigger image of difference being evaluated, even if they have Identical fuzzy quantity, but its evaluation result also differs bigger.How to realize obscure unrelated with picture material Evaluation methodology is the key of push-broom type remote sensing camera Atomatic focusing method.

Summary of the invention

The present invention proposes a kind of push-broom type remote sensing camera Atomatic focusing method based on Wavelet Packet Energy Spectrum, energy It is enough by the result of calculation of remote sensing images motion vertical direction out of focus each frequency range Wavelet Packet Energy Spectrum is carried out Weighted sum, it is achieved for the focusing evaluation of push-broom type remote sensing camera, evaluation methodology has the most accurately Property and monotonicity, and accomplish unrelated with picture material.

The present invention decomposes and energy spectrum Index for Calculation based on image wavelet bag, it is proposed that a kind of push-broom type remote sensing The auto-focusing evaluation methodology of camera, its main thought is:

1, remote sensing images are carried out WAVELET PACKET DECOMPOSITION.

WAVELET PACKET DECOMPOSITION technology, on the basis of wavelet decomposition, will stop the medium-high frequency decomposing in wavelet transformation Section wavelet coefficient continues to decompose, and has that frequency decomposition is fine, effectively represent the advantage of local signal and good Good time-frequency characteristic, can become finer frequency component, then even if image only exists by signal decomposition Small fuzzy quantity difference, utilizes WAVELET PACKET DECOMPOSITION also can carry out effective definition evaluation, thus improves The evaluation performance of focusing evaluation methodology.Meanwhile, WAVELET PACKET DECOMPOSITION image frequency domain power spectrum has invariance, In Wavelet Packet Domain, there is the most consistent energy spectrum curve between different content, it is possible to realize and picture material Unrelated focusing evaluation methodology.

2, choose and sweep, with pushing away, the frequency domain frequency range that camera motion direction is vertical, calculate each frequency range wavelet coefficient Wavelet-packet energy spectrum index, weighted sum obtains final clear evaluation index.

Push-broom type remote sensing camera, due to its special imaging mode, moves the image frequency domain that direction is consistent Component can be affected by certain due to camera motion, causes a certain degree of distortion.Choose and transport with camera The frequency domain frequency range that dynamic direction is vertical, can get rid of the interference of camera motion, accurately reflect out of focus factor and cause Image power spectrum change.Simultaneously, it is contemplated that image defocus blur mainly causes declining of image radio-frequency component Subtracting, the weight coefficient high-frequency band bigger weight coefficient of setting being reduced to low frequency frequency range can more have The response diagram of effect is as out of focus situation.

The present invention comprises the steps:

(1) to input remote sensing images F, (x, y) carries out gray-scale map conversion, and brightness normalized obtains pre- Process image G (x, y).If the remote sensing images of input are coloured image, need first to be converted into gray scale Territory, normalizes to [0 1] interval by gray level image brightness.

(2) (x, y) carries out four layers of WAVELET PACKET DECOMPOSITION to the image G obtained step (1) process, obtains every One layer of each frequency range matrix of wavelet coefficients decomposed.For the image of a size of M × N, utilize two dimension yardstick Function and 2-d wavelet function, image G (x, y) can be expressed as:

(1)

Wherein m, n represent the coordinate position of wavelet packet coefficient matrix,It is two dimensional scaling function, ψ^H(x, y), ψ^V(x, y), ψ^D(x, y) respectively along horizontal edge direction H, vertical edge direction V and right The 2-d wavelet function of linea angulata direction D change.j₀It is to start yardstick arbitrarily,Coefficient defines At yardstick j₀G (x, approximation y)；Coefficient is for j >=j₀Addition of level, vertical and right The details in burnt direction.

Wavelet decomposition all can obtain a low frequency frequency range and three high-frequency band each time, and WAVELET PACKET DECOMPOSITION exists Low frequency component and high fdrequency component are decomposed the most again by each layer, thus is obtained more fine frequency range.

(3) four layers of each frequency-domain small wave coefficient matrix for obtaining in step (2), select in third layer Taking and sweep, with pushing away, 1 lower frequency region frequency range that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C₁； 4 frequency domain frequency ranges, its matrix of wavelet coefficients is defined as C_i(i=2,3,4,5);Institute is chosen in the 4th layer Having to push away and sweep 64 high-frequency domain frequency ranges that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C_j (j=6,7...69).Have chosen 69 frequency ranges altogether for calculating its wavelet-packet energy desired value.

(4) the matrix of wavelet coefficients C of each frequency range that will select in step (3)_k(k=1,2...69) enter Row processes, and is calculated and characterizes and the wavelet-packet energy spectrum index Q of each frequency range in camera motion vertical direction_k (k=1,2...69).According to singular value decomposition method, the matrix of wavelet coefficients C of a certain frequency range_kSingular value Decomposition can be write and do:

$C_k={\stackrel{m\×m}{U}}_k{\stackrel{m\×n}{\mathrm{\Σ}}}_k{\stackrel{n\×n}{V}}_k^{\′}---\left(2\right)$

Wherein C_kIt is the matrix of wavelet coefficients on m × n rank, U_kIt is the unitary matrice of m × m, V_k' it is V_kConjugation Transposition, is the unitary matrice of n × n, U_k′U_k=I, V_k′V_k=I, Σ_kIt is positive semidefinite m × n rank diagonal matrix,

σ_kiIt is referred to as coefficient matrix C_kSingular value, according to descending be:

σ_k1≥σ_k2≥…≥σ_kr＞ 0

σ in theory_k1Much larger than other singular value, available σ_k1Representing matrix C_kMost energy. In order to eliminate the impact on energy of the sample randomness so that with the different scene images under obscuring at same frequency Energy under Duan can be approximately the same, can be selected for σ_k1Represent wavelet-packet energy, and remove other singular values Impact on energy.Wavelet-packet energy spectrum index Q_kIt is expressed as:

Q_k=log₁₀(1+σ_k1) (4)

Wherein, σ_k1Represent the wavelet packet coefficient Matrix C corresponding to a certain frequency range_kCarry out singular value decomposition, Obtain C_kCorresponding diagonal angle eigenvalue matrix, the maximum eigenvalue therefrom chosen.It is carried out logarithm behaviour Make, obtain Wavelet Packet Energy Spectrum exponential quantity Q of this frequency range_k。

(5) to each frequency range Wavelet Packet Energy Spectrum exponential quantity weighted sum chosen in step (4), obtain Final sharpness evaluation function:

$\mathrm{Metric}=\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{Q}_k\×P_k^2---\left(5\right)$

Wherein N=69, the frequency range number selected by expression.K represents the order sequence number of selected frequency range, Q_kTable Show the Wavelet Packet Energy Spectrum exponential quantity of certain frequency range, P_kFor corresponding weight coefficient, P_k=k/N, represents frequency range Order is the highest, and its weight coefficient value is the biggest.Mainly cause the high frequency attenuation of image due to out of focus, thus add The weight coefficient of strong high frequency and reduce the weight coefficient of low frequency and can effectively reflect the out of focus situation of image. This evaluation index evaluation of estimate is the least, shows that image is the fuzzyyest.Camera is carried out according to image definition evaluation value Focusing.

(6) repeat (2)～(5), select the focusing position corresponding to definition evaluation of estimate maximum For final camera blur-free imaging position, terminate auto-focusing.

The present invention utilizes WAVELET PACKET DECOMPOSITION technology, and the remote sensing images that push-broom type remote sensing camera obtains are carried out four Layer WAVELET PACKET DECOMPOSITION, chooses and the frequency domain frequency range of camera motion direction perpendicularity, and calculates each frequency range Wavelet Packet Energy Spectrum exponential quantity, obtains final Image Definition by its weighted sum, in order to Camera is instructed to carry out auto-focusing.Present invention can apply to push-broom type remote sensing camera Autofocus Technology neck Territory, can realize the focusing evaluation unrelated with picture material.

Accompanying drawing explanation

Fig. 1 is the inventive method flow chart.

Fig. 2 is 2-d wavelet bag decomposing schematic representation, and Decomposition order is 2 layers.

Fig. 3 is the schematic diagram of 69 vertical with camera motion direction the frequency domain frequency range chosen.

Fig. 4 (a) is 5 width wavelet-packet energy spectral curve test experiments remote sensing images.

Fig. 4 (b) is 5 width remote sensing figure Wavelet Packet Energy Spectrum Dependence Results.

Fig. 5 is the clear remote sensing figure that 25 width contents are different.

Fig. 6 is 25 width images definition evaluation result under different defocusing amounts in Fig. 5, has reacted right The monotonicity of burnt evaluation algorithms and concordance.

Detailed description of the invention

Below in conjunction with accompanying drawing, the invention will be further described.

The flow chart of the present invention is as it is shown in figure 1, concretely comprise the following steps:

(1) input remote sensing images F is carried out gray-scale map conversion, brightness normalized, obtain pretreatment Image G.

(2) step (1) is processed the image G obtained and carry out four layers of WAVELET PACKET DECOMPOSITION, obtain each layer The each frequency range matrix of wavelet coefficients decomposed.

(3) four layers of each frequency-domain small wave coefficient matrix for obtaining in step (2), select in third layer Taking and sweep, with pushing away, 1 lower frequency region frequency range that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C₁； 4 frequency domain frequency ranges, its matrix of wavelet coefficients is defined as C_i, i=2,3,4,5；Institute is chosen in the 4th layer Having and sweep, with pushing away, 64 high-frequency domain frequency ranges that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C_j, J=6,7...69.

(4) the matrix of wavelet coefficients C of each frequency range that will select in step (3)_kProcess, K=1,2...69, be calculated and characterize and the wavelet-packet energy spectrum index of each frequency range in camera motion vertical direction Q_k；Wavelet-packet energy spectrum index Q_kIt is expressed as:

Q_k=log₁₀(1+σ_k1)

Wherein, σ_k1Represent the wavelet packet coefficient Matrix C corresponding to a certain frequency range_kCarry out singular value decomposition, Obtain C_kCorresponding diagonal angle eigenvalue matrix, the maximum eigenvalue therefrom chosen；It is carried out logarithm behaviour Make, obtain Wavelet Packet Energy Spectrum exponential quantity Q of this frequency range_k。

(5) each frequency range Wavelet Packet Energy Spectrum exponential quantity calculated to step (4), is weighted asking With, obtain sharpness evaluation function；This evaluation index evaluation of estimate is the least, shows that image is the fuzzyyest；According to Definition evaluation of estimate instructs push-broom type remote sensing camera auto-focusing to adjust.

(6) repeat (2)～(5), select the focusing position corresponding to definition evaluation of estimate maximum For final camera blur-free imaging position, terminate auto-focusing.

Fig. 2 is the schematic diagram of the WAVELET PACKET DECOMPOSITION of a two-layer.Each layer decompose can be by operated frequency Section is divided into 4 frequency ranges: a low frequency frequency range, the high-frequency band of a horizontal direction, a vertical direction High-frequency band and the high-frequency band of a diagonal.WAVELET PACKET DECOMPOSITION at each layer not only to low frequency Frequency range is decomposed, and proceeds equally for high-frequency band to decompose, and obtain under each frequency domain components successively is more Fine frequency range.For a certain layer decomposes, raise the most successively according to frequency range shown in schematic diagram.

After algorithm carries out four layers of WAVELET PACKET DECOMPOSITION to image, need to choose the frequency vertical with camera motion direction Territory frequency range is used for calculating wavelet-packet energy spectrum index.Assume that camera motion direction is image level direction, figure The schematic diagram (WAVELET PACKET DECOMPOSITION of some sublayers is the most all listed) that 3 is 69 selected frequency domain frequency ranges, The square frame that wherein WWV represents in third layer is 1 the low frequency frequency range chosen, WVW, WVH, WVV and The square frame that WVD represents is 4 intermediate-frequency bands chosen, and they are the low frequencies in being decomposed by image ground floor Component decomposes further and obtains；The square frame represented by from VWW to VDD in the 4th layer is selected 64 high-frequency band, frequency range raises the most successively, and they are hanging down in being decomposed by image ground floor Nogata decomposes further to details coefficients and obtains.

In order to the invariance feature of wavelet-packet energy power spectrum is described, 5 width test images in Fig. 4 (a), Having neighborhood similarity between them, image scene is not quite identical, including different structure scene altimetric image Content, Ye You city, existing ocean, it is used for simulating the imaging contexts of true push-broom type remote sensing camera.Fig. 4 B in (), solid line represents the wavelet-packet energy spectral curve of this 5 width test picture rich in detail, synteny does not represents Different content image, the dotted line in Fig. 4 (b) represent this 5 width picture rich in detail through a certain amount of identical from Wavelet energy spectral curve after Jiao is fuzzy.As seen from the figure, different images has under identical fog-level The most consistent wavelet-packet energy spectral curve, and during fuzzy increase, high-frequency energy is decayed, curve declines. Can the focusing definition situation of effective response diagram picture with wavelet-packet energy spectrum index.

For monotonicity and the concordance of verification method, Fig. 5 is 25 width test images, comprises difference Content scene, difference is the biggest.To each picture rich in detail, from left to right, it is sequentially added into from top to bottom uniformly The fuzzy out of focus situation different to different scenes in order to emulate remote sensing camera of disk.Blur radius is from 1 picture Element is to 25 pixels, and step-length is 1 pixel, obtains 25 width contents differences and the fuzzy figure increased successively Picture.Processing 25 width out-of-focus images with the algorithm proposed, Fig. 6 is its definition evaluation of estimate, Abscissa is Gaussian Blur standard deviation, and vertical coordinate is the definition evaluation index of each image.It can be seen that For the image of different scenes, focusing evaluation methodology based on Wavelet Packet Energy Spectrum has good monotonicity And concordance, it is possible to effectively instruct push-broom type remote sensing camera auto-focusing.

Claims (2)

1. a push-broom type remote sensing camera Atomatic focusing method based on Wavelet Packet Energy Spectrum, it is characterised in that The method comprises the following steps:

(1) input remote sensing images F is carried out gray-scale map conversion, brightness normalized, obtain pretreatment figure As G；

(2) step (1) is processed the image G obtained and carry out four layers of WAVELET PACKET DECOMPOSITION, obtain each layer The each frequency range matrix of wavelet coefficients decomposed；

(3) four layers of each frequency-domain small wave coefficient matrix for obtaining in step (2), select in third layer Taking and sweep, with pushing away, 1 lower frequency region frequency range that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C₁；4 Individual frequency domain frequency range, its matrix of wavelet coefficients is defined as C_i, i=2,3,4,5；4th layer is chosen all with Pushing away and sweep 64 high-frequency domain frequency ranges that camera motion direction is vertical, its matrix of wavelet coefficients is defined as C_j, J=6,7...69；

(4) the matrix of wavelet coefficients C of each frequency range that will select in step (3)_kProcess, K=1,2...69, be calculated and characterize and the wavelet-packet energy spectrum index of each frequency range in camera motion vertical direction Q_k；Wavelet-packet energy spectrum index Q_kIt is expressed as:

Q_k=log₁₀(1+σ_k1) (1)

Wherein, σ_k1Represent the wavelet packet coefficient Matrix C corresponding to a certain frequency range_kCarry out singular value decomposition, To C_kCorresponding diagonal angle eigenvalue matrix, the maximum eigenvalue therefrom chosen；It is carried out log operations, Obtain Wavelet Packet Energy Spectrum exponential quantity Q of this frequency range_k；

(5) each frequency range Wavelet Packet Energy Spectrum exponential quantity calculated to step (4), is weighted asking With, obtain sharpness evaluation function；

$Metric=\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{Q}_k\×P_k^2---\left(2\right)$

Wherein N=69, the frequency range number selected by expression；K represents the order sequence number of selected frequency range, Q_kTable Show the Wavelet Packet Energy Spectrum exponential quantity of certain frequency range, P_kFor corresponding weight coefficient, P_k=k/N, represents frequency range Order is the highest, and its weight coefficient value is the biggest；Mainly cause the high frequency attenuation of image due to out of focus, thus add The weight coefficient of strong high frequency and reduce the weight coefficient of low frequency and effectively reflect the out of focus situation of image；This is commented Valency metrics evaluation value is the least, shows that image is the fuzzyyest；Push-broom type remote sensing is instructed according to definition evaluation of estimate Camera auto-focusing adjusts；

(6) repeating (2)～(5), selecting the focusing position corresponding to definition evaluation of estimate maximum is Final camera blur-free imaging position, terminates auto-focusing.

A kind of push-broom type remote sensing camera based on Wavelet Packet Energy Spectrum is the most right Burnt method, it is characterised in that: after remote sensing images carry out four layers of WAVELET PACKET DECOMPOSITION, WAVELET PACKET DECOMPOSITION is to low While frequency component decomposes, centering high-frequency wavelet coefficient continues to decompose, and obtains more fine frequency component, Even if there is small fuzzy quantity difference between image, it is also carried out auto-focusing definition evaluation.