CN105957034A - Scanning infrared imaging system scene non-uniformity correction based on registration - Google Patents

Abstract

The invention relates to a scanning infrared imaging system scene non-uniformity correction based on registration, and the method comprises that registration is performed on previous and next frames of video streams, and longitudinal translation amount and transverse translation amount of the screen is calculated. The translation is performed on an image of the previous frame through utilizing Fourier transform operation, and the above image is subtracted by an image of next frame and divides an absolute value of the longitudinal translation amount to obtain a difference matrix. The elementary transformation is performed on the difference matrix to obtain a bias non-uniform difference matrix which aims at an appointed row, and each column of the matrix is partly accumulated to obtain a bias correction matrix. Average solving is performed on each column of the bias correction matrix, so that a bias correction vector is obtained and is used for non-uniform correction of a to-be-corrected image. The method has advantages of fast calculation speed and good quality of image and has good application prospect.

Description

Sweep type infrared imaging system scene asymmetric correction method based on registration

Technical field

The present invention relates to the scene non-uniform correction method of a kind of infrared image, it is more particularly related to one is based on joining Accurate sweep type infrared imaging system scene asymmetric correction method.

Background technology

Infrared imaging system capacity of resisting disturbance is strong, and hidden performance is good, and air penetration capacity is strong, adapts to multiple special occasions, in section Grind, military affairs, medical science, industry, many aspects such as civilian in an increasingly wide range of applications.Now widely used infrared imaging System is divided into two classes: a class is gazing type infrared imaging system, and in system, opticator focuses on infrared focus plane IR Scene On；Another kind of is sweep type infrared imaging system, and system is enough become with alignment infrared focus plane two parts by optical mechaical scanning, system handle Scene is progressively mapped in infrared imaging alignment along scanning direction.

Either in sweep type infrared imaging system or in gazing type infrared imaging system, affected by manufacturing process, The response of each pixel of infrared focus plane is not consistent, there is heterogeneity, shows as fixing pattern noise (fixed in the picture Pattern noise, referred to as FPN), cause infrared image signal to noise ratio low, poor image quality.It is thus desirable to infrared image is carried out Nonuniformity correction processes removes FPN.Owing to FPN exists drift characteristic, self-adapting correction method energy based on scene in time domain Enough it is corrected from heteropical form of expression, the correction error that response drift is brought can be overcome to a certain extent, Do not require or have only to demarcate simply, adaptively updating correction coefficient according to scene information.

At present the most ripe self-adapting correction method based on scene have time domain processing algorithm, spatial processing algorithm and based on The Processing Algorithm of estimation, but the proposition of these methods is both for what gazing type infrared imaging system proposed, and sweep type is red The scene non-uniform correction method of outer imaging system is also not affected by paying attention to.Directly the scene gazing type infrared imaging system is non-homogeneous Bearing calibration applies in sweep type infrared imaging system, there is certain defect: cause convergence rate slow；Do not account for sweeping Retouch the noise behavior of type infrared imaging system, the poor image quality processed can be caused.

Summary of the invention

The present invention is directed to prior art defect, it is provided that a kind of sweep type infrared imaging system scene Nonuniformity Correction based on registration Method.

Technical scheme provides a kind of sweep type infrared imaging system scene asymmetric correction method based on registration, bag Include following steps:

(1) integer constant L more than zero that input is preset, if the resolution of the output image of Infrared Detectors is M × N, makes m, n represent Space coordinates, has m=1 ..., M, n=1 ..., N, the memory space P that size is M × L is set；Counting variable t is set, Line variable s is set, initializes s=1, frame number indexed variable k is set, initialize k=1；

(2) two two field pictures before and after obtaining continuously from the video flowing of Infrared Detectors, are designated as y_kAnd y_k+1；This two two field picture is registrated, Calculate scene translation pixel count, including longitudinal translation d_vWith transverse translation d_h, it may be judged whether 0 < | d_v| < 1,

It is to enter step (4),

Otherwise make k=k+2, two new two field pictures are re-executed step (2)；

(3) to y_kCalculate displacement images y_k ^*It is as follows,

$y_k^{*}(m,n)=IFFT\left(Y_k\right(u,v\left)\exp \right(-j2\mathrm{\&pi;}\left(\frac{{\mathrm{ud}}_v}{M}+\frac{{\mathrm{vd}}_h}{N}\right)\left)\right)$

Wherein, m, n are image space coordinate, and u, v are frequency domain coordinate, Y_k(u v) is y_kFrequency domain figure picture,It is y_k Displacement images, j is imaginary unit；

Calculate difference matrix D as follows,

$D=\frac{1}{\left|d_v\right|}(y_k^{*}-y_{k+1})$

If d_v> 0, the first row of D is set to 0；If d_v< 0, last column of D is set to 0；

D_hIt is divided into integer part Int (d_h) and fractional part ε (d_h), i.e. d_h=Int (d_h)+ε(d_h), if d_h> 0, remove row in D and sit It is designated as n=1 ..., 1+ | Int (d_h) | row；If d_h< 0, removing row coordinate in D is n=N-| Int (d_h) | ..., the row of N；

(4) difference matrix D is carried out elementary transformation, obtain biasing heteropical difference matrix G for s row；

For d_v> 0 situation, if s=1, it is not necessary to perform any operation；If 2≤s≤M, from 1 row to s-1 row, sequentially The element of current line is set to the opposite number of next line element by ground；

For d_v< situation of 0, if s=M, it is not necessary to would perform any operation；If 1≤s≤M-1, inverted order ground is by the element of current line It is set to the opposite number of lastrow element；

(5) perform to add up to each column in G, obtain biasing heteropical correction matrix Q for s row, each in correction matrix Q Element computational methods are as follows,

As m=s, make Q (m, n)=0；

Work as m=1 ..., during s-1,

$Q(m,n)=\underset{r=m}{\overset{s-1}{\&Sigma;}}G(r,n),n=1,\mathrm{...},N-\left|Int\left(d_h\right)\right|-1$

Work as m=s+1 ..., during M,

$Q(m,n)=\underset{r=s+1}{\overset{m}{\&Sigma;}}G(r,n),n=1,\mathrm{...},N-\left|Int\left(d_h\right)\right|-1$

(6) Q is averaging by row, obtains the updating vector c that dimension is M × 1^s, m row element computing formula is as follows,

$c^s(m,1)=\frac{1}{N-\left|Int\left(d_h\right)\right|-1}\underset{n=1}{\overset{N-\left|Int\left(d_h\right)\right|-1}{\&Sigma;}}Q(m,n)$

Updating vector c^sValue be saved in memory space P t row in；

(7) if s < M, s=s+1 is made；Otherwise, s=1 is made；

(8) if t < L, make t=t+1, make k=k+2, return step (2) and proceed new two continuous frames to process；Otherwise, to storage Space P is averaging by row, obtains average correction vector c_avg；

(9) average correction vector c is used_avgThe output image of Infrared Detectors is carried out Nonuniformity Correction；

(10) make t=1, memory space P is reset；

(11) make k=k+2, return step (2) and new two continuous frames is processed, dynamically update average correction vector c_avg。

And, the described opposite number that sequentially element of current line is set to next line element, it is achieved as follows,

D (m, n)=-D (m+1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to 1,2 ..., s-1, gradually add 1.

And, the element of current line is set to the opposite number of lastrow element by described inverted order ground, it is achieved as follows,

D (m, n)=-D (m-1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to M, M-1 ..., s+1, gradually subtract 1；Finally the element of s row is set to 0, by the square after conversion Battle array is designated as G.

The scene asymmetric correction method of the present invention has the advantage that

1. the amount of calculation needed for is little, and resource consumption is few.

2. calculation is succinct, it is easy to accomplish.

3. the image obtained after correction is clear, and uniformity is good.

Accompanying drawing explanation

Fig. 1 is the method flow diagram of the embodiment of the present invention.

Detailed description of the invention

For the ease of those of ordinary skill in the art understand and implement the present invention, below in conjunction with the accompanying drawings and embodiment the present invention is made into The detailed description of one step, it will be appreciated that embodiment described herein is merely to illustrate and explains the present invention, is not used to limit The present invention.

When being embodied as, method provided by the present invention can use computer software technology to realize automatically and run.As it is shown in figure 1, this Inventive embodiments sweep type infrared imaging system scene asymmetric correction method based on registration comprises the following steps:

(1) input integer constant L more than zero, when being embodied as, this value can be by those skilled in the art according to heteropical noise Level-preset, such as, arrange L=10 or L=20, etc..If the resolution of the output image of Infrared Detectors is M × N, order M, n representation space coordinate, has m=1 ..., M, n=1 ..., N, the memory space P that size is M × L is set.Counting is set Variable t.Line variable s is set, initializes s=1.Frame number indexed variable k is set, initializes k=1；

(2) two two field pictures before and after obtaining continuously from the video flowing of Infrared Detectors, are designated as y_kAnd y_k+1.This two two field picture is registrated, Calculate scene translation pixel count, including longitudinal translation d_vWith transverse translation d_h, to there is the image of sub-pix longitudinal translation to entering Row subsequent treatment, i.e. exist sub-pix longitudinal translation (0 < | d_v| < 1, d_hCan be arbitrarily).When being embodied as, it can be determined that whether 0<|d_v| < 1, it is to enter step (4), otherwise make k=k+2, two new frames are re-executed step (2).

(3) to y_kCarry out fast Fourier transform (FFT) and obtain its frequency domain figure as Y_k, by Y_kIt is multiplied by with translation information d_vAnd d_hRelevant Complex exponential, then carry out an inverse fast fourier (IFFT) and obtain y_kDisplacement images y_k ^*, i.e.

$y_k^{*}(m,n)=IFFT\left(Y_k\right(u,v\left)\exp \right(-j2\mathrm{\&pi;}\left(\frac{{\mathrm{ud}}_v}{M}+\frac{{\mathrm{vd}}_h}{N}\right)\left)\right)$

Wherein m, n are image space coordinate, and u, v are frequency domain coordinate, i.e. Y_k(u v) is y_kFrequency domain figure picture,It is y_k Displacement images, j is imaginary unit.

Again y_k ^*With y_k+1Subtract each other and divided by | d_v|, obtaining difference matrix D, computing formula is

$D=\frac{1}{\left|d_v\right|}(y_k^{*}-y_{k+1})$

If d_v> 0, the first row of D is set to 0；If d_v< 0, last column of D is set to 0；

D_hIt is divided into integer part Int (d_h) and fractional part ε (d_h), i.e. d_h=Int (d_h)+ε(d_h), if d_h> 0, remove row in D and sit It is designated as n=1 ..., 1+ | Int (d_h) | row；If d_h< 0, removing row coordinate in D is n=N-| Int (d_h) | ..., the row of N.

(4) difference matrix D is carried out elementary transformation, obtain biasing heteropical difference matrix G for s row.

For d_v> 0 situation, if s=1, it is not necessary to perform any operation；If 2≤s≤M, from 1 row to s-1 row, sequentially The element of current line is set to the opposite number of next line element by ground, it may be assumed that

D (m, n)=-D (m+1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to 1,2 ..., s-1, gradually add 1；

For d_v< situation of 0, if s=M, it is not necessary to would perform any operation；If 1≤s≤M-1, inverted order ground is by the element of current line It is set to the opposite number of lastrow element, it may be assumed that

D (m, n)=-D (m-1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to M, M-1 ..., s+1, gradually subtract 1；Finally the element of s row is set to 0, by the square after conversion Battle array is designated as G.

(5) perform to add up to each column in G, thus obtain biasing heteropical correction matrix Q for s row；In matrix Q respectively Element computational methods are as follows:

As m=s, make Q (m, n)=0；

Work as m=1 ..., during s-1,

$Q(m,n)=\underset{r=m}{\overset{s-1}{\&Sigma;}}G(r,n),n=1,\mathrm{...},N-\left|Int\left(d_h\right)\right|-1$

Work as m=s+1 ..., during M,

$Q(m,n)=\underset{r=s+1}{\overset{m}{\&Sigma;}}G(r,n),n=1,\mathrm{...},N-\left|Int\left(d_h\right)\right|-1$

(6) Q is averaging by row, obtains the updating vector c that dimension is M × 1^s, wherein m row (m=1 ..., M) element computing formula As follows:

$c^s(m,1)=\frac{1}{N-\left|Int\left(d_h\right)\right|-1}\underset{n=1}{\overset{N-\left|Int\left(d_h\right)\right|-1}{\&Sigma;}}Q(m,n)$

Updating vector c^sValue be saved in memory space P t row in, i.e.

P (m, t)=c^s(m,1)

(7) if s < M, make s from increasing 1, i.e. s=s+1；Otherwise, s=1 is made.The most often calculate once, make s=s+1, if s+1 > M, Then make s=1.

(8) if t < L, make t from increasing 1, i.e. t=t+1, make k=k+2, return step (2) and proceed new two continuous frames to process； Otherwise, then memory space P is averaging by row, obtains average correction vector c_avg, i.e.

$c_{avg}(m,1)=\frac{1}{L}\underset{l=1}{\overset{L}{\&Sigma;}}P(m,l),m=1,...,M$

(9) average correction vector c is used_avgThe output image of Infrared Detectors is carried out Nonuniformity Correction, and updating formula is as follows:

y_c(m, n)=y (m, n)+c_avg(m,1)

Wherein m=1 ..., M, n=1 ..., N, y are the image to be corrected of M × N, y_cFor image after the correction of M × N.

(10) t=1 is made.Memory space P is reset, i.e.

P (m, l)=0

Wherein, l is the row coordinate of P, l=1 ..., L.

(11) make k=k+2, return step (2) and new two continuous frames is processed, thus dynamically update average correction vector c_avg, with Just heteropical time drift problem is solved, until stopping flow process.

Using method provided by the present invention to test the emulating image with fixing pattern noise, this image is in video flowing One frame, due to the time drift characteristic of fixing pattern noise, causes nonuniformity correction parameter undesirable, therefore exists big in image Measure horizontal stripe noise as depicted.Design sketch after processing from using the inventive method, it is found that compare figure before treatment Picture, horizontal stripe noise substantially disappears, it is possible to the detailed information being clearly observed in scene.As can be seen here, the present invention proposes Method can improve heterogeneity effectively.

Claims (3)

1. a sweep type infrared imaging system scene asymmetric correction method based on registration, it is characterised in that comprise the following steps:

(1) integer constant L more than zero that input is preset, if the resolution of the output image of Infrared Detectors is M × N, makes m, n represent Space coordinates, has m=1 ..., M, n=1 ..., N, the memory space P that size is M × L is set；Counting variable t is set, Line variable s is set, initializes s=1, frame number indexed variable k is set, initialize k=1；

(2) two two field pictures before and after obtaining continuously from the video flowing of Infrared Detectors, are designated as y_kAnd y_k+1；This two two field picture is registrated, Calculate scene translation pixel count, including longitudinal translation d_vWith transverse translation d_h, it may be judged whether 0 < | d_v| < 1,

It is to enter step (4),

Otherwise make k=k+2, two new two field pictures are re-executed step (2)；

(3) to y_kCalculate displacement images y_k ^*It is as follows,

$y_k^{*}(m,n)=IFFT\left(Y_k\right(u,v\left)\exp \right(-j2\mathrm{\&pi;}\left(\frac{{\mathrm{ud}}_v}{M}+\frac{{\mathrm{vd}}_h}{N}\right)\left)\right)$

Wherein, m, n are image space coordinate, and u, v are frequency domain coordinate, Y_k(u v) is y_kFrequency domain figure picture,It is y_k Displacement images, j is imaginary unit；

Calculate difference matrix D as follows,

$D=\frac{1}{\left|d_v\right|}(y_k^{*}-y_{k+1})$

If d_v> 0, the first row of D is set to 0；If d_v< 0, last column of D is set to 0；

D_hIt is divided into integer part Int (d_h) and fractional part ε (d_h), i.e. d_h=Int (d_h)+ε(d_h), if d_h> 0, remove row in D and sit It is designated as n=1 ..., 1+ | Int (d_h) | row；If d_h< 0, removing row coordinate in D is n=N-| Int (d_h) | ..., the row of N；

(4) difference matrix D is carried out elementary transformation, obtain biasing heteropical difference matrix G for s row；

For d_v> 0 situation, if s=1, it is not necessary to perform any operation；If 2≤s≤M, from 1 row to s-1 row, sequentially The element of current line is set to the opposite number of next line element by ground；

For d_v< situation of 0, if s=M, it is not necessary to would perform any operation；If 1≤s≤M-1, inverted order ground is by the element of current line It is set to the opposite number of lastrow element；

(5) perform to add up to each column in G, obtain biasing heteropical correction matrix Q for s row, each in correction matrix Q Element computational methods are as follows,

As m=s, make Q (m, n)=0；

Work as m=1 ..., during s-1,

$Q(m,n)=\underset{r=m}{\overset{s-1}{\mathrm{\&Sigma;}}}G(r,n),n=1,...,N-\left|Int\left(d_h\right)\right|-1$

Work as m=s+1 ..., during M,

$Q(m,n)=\underset{r=s+1}{\overset{m}{\mathrm{\&Sigma;}}}G(r,n),n=1,...,N-\left|Int\left(d_h\right)\right|-1$

(6) Q is averaging by row, obtains the updating vector c that dimension is M × 1^s, m row element computing formula is as follows,

$c^s(m,1)=\frac{1}{N-\left|Int\left(d_h\right)\right|-1}\underset{n=1}{\overset{N-\left|Int\left(d_h\right)\right|-1}{\&Sigma;}}Q(m,n)$

Updating vector c^sValue be saved in memory space P t row in；

(7) if s < M, s=s+1 is made；Otherwise, s=1 is made；

(8) if t < L, make t=t+1, make k=k+2, return step (2) and proceed new two continuous frames to process；Otherwise, to storage Space P is averaging by row, obtains average correction vector c_avg；

(9) average correction vector c is used_avgThe output image of Infrared Detectors is carried out Nonuniformity Correction；

(10) make t=1, memory space P is reset；

(11) make k=k+2, return step (2) and new two continuous frames is processed, dynamically update average correction vector c_avg。

Sweep type infrared imaging system scene asymmetric correction method based on registration the most according to claim 1, it is characterised in that: The described opposite number that sequentially element of current line is set to next line element, it is achieved as follows,

D (m, n)=-D (m+1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to 1,2 ..., s-1, gradually add 1.

Sweep type infrared imaging system scene asymmetric correction method based on registration the most according to claim 1, it is characterised in that: The element of current line is set to the opposite number of lastrow element by described inverted order ground, it is achieved as follows,

D (m, n)=-D (m-1, n), n=1 ..., N-| Int (d_h)|-1

Wherein row-coordinate m be changed to M, M-1 ..., s+1, gradually subtract 1；Finally the element of s row is set to 0, by the square after conversion Battle array is designated as G.