CN105005967A - Method and apparatus for correcting non-uniformity of infrared imaging based on combined space-time filtering - Google Patents

Abstract

The present invention discloses a method for correcting non-uniformity of infrared imaging based on combined space-time filtering, comprising the steps of: performing global filtering for an inputted original infrared non-uniformity image x(n) to obtain a globally filtered image z(n); performing subtraction between the original infrared non-uniformity image x(n) and the globally filtered image z(n) to obtain a noise image zG(n); performing filtering processing for the obtained noise image zG(n) to obtain a pure non-uniformity noise image zGr(n) at a low frequency in time domain; and performing subtraction between the original infrared non-uniformity image x(n) and the obtained pure non-uniformity noise image zGr(n) at the low frequency in time domain to obtain a non-uniformity corrected infrared image y(n). The invention also discloses an apparatus for correcting non-uniformity of infrared imaging based on combined space-time filtering. According to the method and the apparatus of the invention, a high level of filtering performance can be kept for a complex image having a local non-repetitive or semi-random structure and an extremely low signal to noise ratio.

Description

The infrared imaging asymmetric correction method that spatio-temporal filtering combines and device thereof

Technical field

The invention belongs to the Nonuniformity Correction field in infrared imaging pre-service, be specifically related to infrared imaging asymmetric correction method that a kind of spatio-temporal filtering combines and device thereof.

Background technology

Infrared focal plane array (Infrared focal plane arrays, IRFPA) panel detector structure is simple, and detection sensitivity is high, and system works frame frequency is high.But by the impact of the factors such as material, manufacturing process and working environment, under homogeneous radiation is irradiated, in infrared focal plane array, between each probe unit, response exports inconsistent, and this inconsistency is called as heterogeneity (Non-uniformity).This heterogeneity causes the temperature resolution of system to decline, the quality of target image is had a strong impact on, it shows as certain fixing pattern usually on image, is therefore usually called as again fixed pattern noise (Fixed Pattern Noise, FPN).Thus in order to improve the image quality of infrared focus plane thermal imaging system, just Nonuniformity Correction process must be carried out to its image.

Non-uniformity Correction Algorithm has multiple, totally can be divided into two large classes.Wherein a class is the Non-uniformity Correction Algorithm based on calibration, and another kind of is Non-uniformity Correction Algorithm based on scene.Non-uniformity Correction Algorithm development based on calibration is very ripe, and actual engineering uses this type of algorithm mostly.But such algorithm needs Periodic calibration, thus need interrupt system normally to work, and destroy requirement of real-time.Correcting algorithm based on scene is a kind of Adaptive correction algorithm.Such algorithm, from heteropical form of expression, utilizes the scene information of real time imagery to upgrade correction coefficient adaptively, can reduce the correction error that probe unit response drift is brought to a certain extent; Such algorithm does not require or only needs to calibrate simply in addition, and does not need the normal work stopping detection system, and realize system succinct, therefore such algorithm becomes the study hotspot of Nonuniformity Correction.

Wherein, the most classical self-adaptation nonuniformity correction technology based on scene mainly contains following two kinds: one is temporal high pass filter correcting algorithm (is called for short THPF-NUC), and also having a kind of is take neural computing as the adaptive nonuniformity correction algorithm (being called for short NNC) of core.THPF algorithm is based on following consideration: the response characteristic of infrared focal plane array probe unit changes slowly in time, and therefore heterogeneity can be considered constant at short notice, namely belongs to the low-frequency component in time domain; And relative motion faster target scene belong to the radio-frequency component in time domain.Therefore, the heterogeneity noise of temporal high pass filter device filtering low frequency can be utilized and the scene of reserved high-frequency.But the method requires that scene is ceaselessly moved, when scene stillness or when obviously accelerating, correction parameter is affected very greatly, the generation of objective fuzzy and " ghost " phenomenon can be caused, reduce convergence of algorithm speed.NNC algorithm utilizes the result of study that there is lateral ties and backfeed loop between amphiblestroid visual cell, adopt neural network structure, the expectation value exported using four neighboring mean values as pixel, and adopt steepest descent method to upgrade correction coefficient iteration, finally realize the adaptively correcting to IRFPA.This algorithm can the correction of complete pair gain and offset response coefficient, relatively better to spatial high-frequency noise suppression effect, but there is target fade-out and " ghost " phenomenon equally, and due to its calculated amount large, speed of convergence is slow, is often difficult to Project Realization.

Summary of the invention

In view of this, the fundamental purpose of the present invention infrared imaging asymmetric correction method that is to provide a kind of spatio-temporal filtering to combine and device thereof

For achieving the above object, technical scheme of the present invention is achieved in that

The infrared imaging asymmetric correction method that the embodiment of the present invention provides a kind of spatio-temporal filtering to combine, the method is: carry out global filtering to original infrared heterogeneity image x (n) of input and obtain image z (n) after global filtering, obtains noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^gn (), to the noise image z of described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^grn (), finally by original infrared heterogeneity image x (n) and the pure heterogeneity noise image z on the temporal low frequency obtained ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

In such scheme, describedly obtain noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^gn (), is specially: z ^g(n)=x (n)-z (n), described noise image z ^ga large amount of heterogeneities and common random noise is contained in (n), and a small amount of scene image details.

In such scheme, to the noise image z of described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^grn (), is specially: by noise image z ^gn (), as the input of IIR type temporal low pass filter, obtains the pure heterogeneity noise image z that the low frequency in time domain exports after filtering process ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

In such scheme, finally by original infrared heterogeneity image x (n) and the heterogeneity image z in the time domain obtained ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction, be specially: y (n)=x (n)-z ^gr(n).

The infrared imaging Nonuniformity Correction device that the embodiment of the present invention also provides a kind of spatio-temporal filtering to combine, this device comprises global filtering unit, noise image obtains unit, temporal low frequency filter unit, Nonuniformity Correction unit;

Described global filtering unit, obtains image z (n) after global filtering for carrying out global filtering to original infrared heterogeneity image x (n) of input;

Described noise image obtains unit, for obtaining noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^g(n);

Described temporal low frequency filter unit, for the noise image z to described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^gr(n);

Described Nonuniformity Correction unit, for by original infrared heterogeneity image x (n) with obtain temporal low frequency on pure heterogeneity noise image z ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

In such scheme, described noise image obtains unit, specifically for according to z ^gn ()=x (n)-z (n) obtains noise image and obtains unit, described noise image z ^ga large amount of heterogeneities and common random noise is contained in (n), and a small amount of scene image details.

In such scheme, described temporal low frequency filter unit, specifically for by noise image z ^gn (), as the input of IIR type temporal low pass filter, obtains the pure heterogeneity noise image z that the low frequency on temporal low frequency exports after filtering process ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

In such scheme, described Nonuniformity Correction unit, specifically for according to y (n)=x (n)-z ^grn () obtains infrared image y (n) after Nonuniformity Correction.

Compared with prior art, beneficial effect of the present invention:

The present invention not only possesses the advantage of the edge-protected characteristic of spatial filter; and to there being the complicated image that the signal to noise ratio (S/N ratio) of local non-repeatability or half random structure is very low; its filtering performance still keeps higher level; in addition; due to only use the remaining noise image of global filtering (wherein only containing heterogeneity and common high frequency noise) during fine correction but not former figure as the input of temporal low pass filter; and almost eliminate the interference of scene information; therefore avoid the appearance of " ghost " phenomenon, convergence of algorithm speed is significantly improved.

Accompanying drawing explanation

The process flow diagram of the infrared imaging asymmetric correction method that Fig. 1 provides a kind of spatio-temporal filtering to combine for the embodiment of the present invention;

Fig. 2 is the correction results contrast of different bearing calibration.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

The infrared imaging asymmetric correction method that the embodiment of the present invention provides a kind of spatio-temporal filtering to combine, as shown in Figure 1, the method is realized by following steps:

Step 101: global filtering is carried out to original infrared heterogeneity image x (n) of input and obtains image z (n) after global filtering.

Concrete, input original infrared heterogeneity image x (n), carry out global filtering to it and obtain image z (n) after global filtering, wherein n represents the n-th two field picture.

The overall spatial filter that global filtering is selected is a kind of nonlinear filter, and this wave filter not only protects the edge details of image, and still fine to the complicated image filtering performance that signal to noise ratio (S/N ratio) is very low.Therefore the scene information only containing a large amount of heterogeneities and random noise component not containing complexity in the noise image obtained, effectively can avoid the interference of scene information, suppress the generation of " ghost " phenomenon, greatly increase convergence of algorithm speed simultaneously.

Step 102: obtain noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^g(n).

Concrete, according to z ^gn ()=x (n)-z (n) obtains noise image z ^g(n), described noise image z ^gn () contains a large amount of heterogeneities and common random noise, and a small amount of scene image details.

Step 103: to the noise image z of described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^gr(n).

Concrete, by noise image z ^gn (), as the input of IIR type temporal low pass filter, the low frequency obtained after filtering process in time domain exports z ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

Adopt IIR type temporal low pass filter to noise image z ^gn () carries out iteration frame by frame, reduce the complexity of algorithm, and be convenient to engineer applied.

Step 104: finally by original infrared heterogeneity image x (n) and the pure heterogeneity noise z on the temporal low frequency obtained ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

Concrete, the pure heterogeneity noise image z exported by original heterogeneity infrared image x (n) and low pass obtained in the previous step ^grn () subtracts each other, namely obtain infrared image y (n) after Nonuniformity Correction, and formula is: y (n)=x (n)-z ^gr(n).

First the present invention adopts nonlinear overall spatial filter slightly to correct real heterogeneity infrared image, then adopts temporal low pass filter iteration to carry out heteropical fine correction; The present invention not only possesses the advantage of the edge-protected characteristic of spatial filter; and to there being the complicated image that the signal to noise ratio (S/N ratio) of local non-repeatability or half random structure is very low; its filtering performance still keeps higher level; in addition; due to only use the remaining noise image of global filtering (wherein only containing heterogeneity and common high frequency noise) during fine correction but not former figure as the input of temporal low pass filter; and almost eliminate the interference of scene information; therefore avoid the appearance of " ghost " phenomenon, convergence of algorithm speed is significantly improved.

Emulation experiment

1, experiment condition

All experiments herein are all configured to the internal memory of CPU, 4GB of the 1.70GHz of i5 at one, the computer platform of the 64 bit manipulation systems of Windows7, by Matlab2010b software platform programming realization.200 two field pictures that the infrared image sequence that this experiment adopts is taken with the frame frequency of 25 frames/second by non-brake method 256 × 256 pixel IRFPA camera form.During experiment, 180 × 200 pixel regions only having intercepted heterogeneity in the middle part of image larger calculate.In experiment, parameter A is set to 3, to the image before and after the correction of the 129th frame heterogeneity image by shown in Fig. 2 (a) ~ figure (d).

2, experiment content

The correction result figure that Fig. 2 provides the 129th frame heterogeneity image and obtains with diverse ways.Wherein, a () is the real infrared heterogeneity image of the 129th frame, b () be rear image for THPF method corrects, (c) is image after NNC method corrects, and (d) is image after the inventive method (GLC) corrects.Relatively five width images can be found out, (a) thickens, although eliminate heterogeneity striped, image also serious degradation by the image after THPF method corrects owing to being covered by a large amount of vertical heterogeneity stripeds; Image after being corrected by NNC method, still containing a large amount of heterogeneity striped, corrects result also unsatisfactory; And after adopting the inventive method to correct, the heterogeneity striped of image is almost all removed, and not containing obvious " ghost " phenomenon, image is more clear.

In objective evaluation index, heterogeneity (NU) and these two kinds of evaluation indexes of roughness (R) are adopted to compare the correcting feature of the inventive method (GLC) and NNC method and THPF method.The less explanation calibration result of heterogeneity index (NU) is better; Roughness index (R) is relatively less, also illustrates that calibration result is better.

Wherein, heterogeneity (NU) is defined as: under even incident radiation condition, the percentage of the toaverage ratio of the response output of the equivalent signal intensity pixel all with it of the response output of infrared focal plane array seeker pixel.

$\mathrm{NU}=\frac{1}{V_{\mathrm{avg}}}\sqrt{\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{(V_{\mathrm{ij}}-V_{\mathrm{avg}})}^2}$

In formula, M and N is respectively line number and the columns of infrared focal plane array, V _ijfor the response output signal amplitude of the i-th row jth row pixel in focal plane arrays (FPA); V _avgthe mean value of all imaging pixels response output signal amplitude, namely $V_{\mathrm{avg}}=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{V}_{\mathrm{ij}}.$

The computing formula of roughness is,

$\mathrm{\ρ}\left(f\right)=\frac{{\left|\right|h_1*f\left|\right|}_1+{\left|\right|h_2*f\left|\right|}_1}{{\left|\right|f\left|\right|}_1}$

In formula, f represents infrared image to be weighed; h ₁=[1 ,-1], h ₂=[1 ,-1] ^t, both are transposed matrix each other, wherein h ₁represent horizontal direction pattern matrix, h ₂represent vertical direction pattern matrix; " * " expression carries out convolution operation to image; || || ₁represent L ₁norm.

The performance index of distinct methods are as shown in table 1, and table 1 is the Performance comparision of different asymmetric correction method:

From table 1, no matter these two kinds of objective performance index are heterogeneity or roughness, adopt GLC method of the present invention to the Nonuniformity Correction result of image, are all obviously better than NNC and THPF method (wherein NaN represents infinitely great).This also just shows that the inventive method can carry out Nonuniformity Correction to image effectively, and achieves good Nonuniformity Correction effect.

Above-mentioned experiment proves, carry out Nonuniformity Correction process by the present invention to infrared imaging image, the speed of convergence of correction significantly improves, and greatly reduces heterogeneity and the roughness of infrared image, making obtained result images more clear, is a kind of effective asymmetric correction method; No matter be in visual effect or in objective evaluation index, the infrared imaging asymmetric correction method (GLC) that the spatio-temporal filtering that the present invention proposes combines all is better than the classical correcting algorithm of THPF and NNC two kinds.

The embodiment of the present invention additionally provides the infrared imaging Nonuniformity Correction device that a kind of spatio-temporal filtering combines, and this device comprises global filtering unit, noise image obtains unit, temporal low frequency filter unit, Nonuniformity Correction unit;

Described global filtering unit, obtains image z (n) after global filtering for carrying out global filtering to original infrared heterogeneity image x (n) of input;

Described noise image obtains unit, for obtaining noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^g(n);

Described temporal low frequency filter unit, for the noise image z to described acquisition ^gn () obtains the pure heterogeneity noise image z in time domain after carrying out filtering process ^gr(n);

Described Nonuniformity Correction unit, for by original infrared heterogeneity image x (n) with obtain time domain on pure heterogeneity noise image z ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

Described noise image obtains unit, specifically for according to z ^gn ()=x (n)-z (n) obtains noise image and obtains unit, described noise image z ^ga large amount of heterogeneities and common random noise is contained in (n), and a small amount of scene image details.

Described temporal low frequency filter unit, specifically for by noise image z ^gn (), as the input of IIR type temporal low pass filter, obtains the heterogeneity image z that the low frequency in time domain exports after filtering process ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

Described Nonuniformity Correction unit, specifically for according to y (n)=x (n)-z ^grn () obtains infrared image y (n) after Nonuniformity Correction.

The above, be only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.

Claims (8)

1. the infrared imaging asymmetric correction method that combines of a spatio-temporal filtering, it is characterized in that, the method is: carry out global filtering to original infrared heterogeneity image x (n) of input and obtain image z (n) after global filtering, obtains noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^gn (), to the noise image z of described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^grn (), finally by original infrared heterogeneity image x (n) and the pure heterogeneity noise image z on the temporal low frequency obtained ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

2. the infrared imaging asymmetric correction method that combines of spatio-temporal filtering according to claim 1, it is characterized in that, describedly obtaining noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^gn (), is specially: z ^g(n)=x (n)-z (n), described noise image z ^ga large amount of heterogeneities and common random noise is contained in (n), and a small amount of scene image details.

3. the infrared imaging asymmetric correction method that combines of spatio-temporal filtering according to claim 1 and 2, is characterized in that, to the noise image z of described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^grn (), is specially: by noise image z ^gn (), as the input of IIR type temporal low pass filter, obtains the pure heterogeneity noise image z that the low frequency in time domain exports after filtering process ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

4. the infrared imaging asymmetric correction method that combines of spatio-temporal filtering according to claim 3, is characterized in that, finally by original infrared heterogeneity image x (n) and the heterogeneity image z in the time domain obtained ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction, be specially: y (n)=x (n)-z ^gr(n).

5. the infrared imaging Nonuniformity Correction device that combines of spatio-temporal filtering, is characterized in that, this device comprises global filtering unit, noise image obtains unit, temporal low frequency filter unit, Nonuniformity Correction unit;

Described global filtering unit, obtains image z (n) after global filtering for carrying out global filtering to original infrared heterogeneity image x (n) of input;

Described noise image obtains unit, for obtaining noise image z by subtracting each other with image z (n) after original infrared heterogeneity image x (n) and described acquisition global filtering ^g(n);

Described temporal low frequency filter unit, for the noise image z to described acquisition ^gn () obtains the pure heterogeneity noise image z on temporal low frequency after carrying out filtering process ^gr(n);

Described Nonuniformity Correction unit, for by original infrared heterogeneity image x (n) with obtain temporal low frequency on pure heterogeneity noise image z ^grn () subtracts each other infrared image y (n) after obtaining Nonuniformity Correction.

6. the infrared imaging Nonuniformity Correction device that combines of spatio-temporal filtering according to claim 5, is characterized in that: described noise image obtains unit, specifically for according to z ^gn ()=x (n)-z (n) obtains noise image and obtains unit, described noise image z ^ga large amount of heterogeneities and common random noise is contained in (n), and a small amount of scene image details.

7. the infrared imaging Nonuniformity Correction device that the spatio-temporal filtering according to claim 5 or 6 combines, is characterized in that: described temporal low frequency filter unit, specifically for by noise image z ^gn (), as the input of IIR type temporal low pass filter, obtains the pure heterogeneity noise image z that the low frequency on temporal low frequency exports after filtering process ^gr(n), concrete formula is:

$z^{\mathrm{Gr}}\left(n\right)=\frac{1}{A}\·G\·z^G\left(n\right)+(1-\frac{1}{A})\·z^{\mathrm{Gr}}(n-1)$

Wherein, wherein A is the time constant preset, and G is gain compensation factor.

8. the infrared imaging Nonuniformity Correction device that combines of spatio-temporal filtering according to claim 7, is characterized in that: described Nonuniformity Correction unit, specifically for according to y (n)=x (n)-z ^grn () obtains infrared image y (n) after Nonuniformity Correction.