CN104268844A - Small target infrared image processing method based on weighing local image entropy - Google Patents

Abstract

In order to effectively process small target infrared images under the low signal to noise ratio complex background, the invention discloses a small target infrared image processing method based on the weighing local image entropy and relates to the technical field of digital image processing. Inherent features of the small target infrared images are utilized, a multi-scale gray difference operator and a local image entropy operator are provided, the weighing local image entropy is obtained through the dot product operation, so that the infrared image background and the noise are effectively restrained, a target is enhanced, and finally the signal to noise ratio of the images is greatly improved.

Description

A kind of small target infrared image disposal route based on weighting topography entropy

Technical field

The present invention relates to digital image processing techniques field, specifically a kind of small target infrared image disposal route based on weighting topography entropy.

Background technology

Small target infrared image treatment technology civil area (as satellite atmosphere infrared cloud image analyze, Infrared Therapy image pathological analysis, geological analysis, sea personnel search and rescue, intrusion detection, forest fire detect) and military field (as precise guidance, early warning detection, battlefield commander and scouting, enemy and we identify) be used widely, its target detection step is the Focal point and difficult point in infrared image processing field, its performance quality directly determines the EFFECTIVE RANGE of infrared system and the complexity of equipment, thus the research of this technology receives lot of domestic and foreign scholar and continues and general concern.

Target in Small object image is little, intensity is weak, does not have the size of priori, shape and Texture eigenvalue, and together with target, background be aliasing in noise, is difficult to direct-detection.But background it is generally acknowledged to have correlativity on spatial domain, time domain has stability, and be in the low frequency part of image on frequency domain, and target it has been generally acknowledged that on spatial domain uncorrelated with background, and frequency domain is in the HFS of image.Therefore, small target infrared image Processing Algorithm is mainly divided into time domain, spatial domain and transform domain three class: Time-Domain algorithm is mainly used in suppressing to have the background of short-term stationarity, but undesirable to the inhibition of complex background.Air space algorithm has good real-time, is easy to realize.Medium filtering is only suitable for the random noise that elimination pulse width is less than filter window, cannot process structurized background; Top cap conversion is a kind of non-linear background filtering technique of practicality, but needs the priori of image, and adaptivity is not strong; Auto-adaptive filtering technique, as two-dimentional least mean-square error filtering scheduling algorithm, requires that the statistical property of background is constant or slowly change, so effectively cannot suppress complex background.Transform-domain algorithm is as based on the Butterworth high-pass filtering of adaptive frequency territory, wavelet transformation etc., but this type of algorithm derives from Fourier conversion, by the restriction (namely the product of time window and frequency window is a constant) of Heisenberg (Heisenberg) uncertainty principle, and need positive and negative twice conversion, algorithm operation quantity is large.

Although small target infrared image process field has achieved a lot of achievement, and existing a lot of algorithm obtains good realization in engineer applied, but for low signal-to-noise ratio small target infrared image under complex background, its object detection system engineering still faces very large difficulty and complicacy.How to design the key issue that structure is simple, the small target infrared image Processing Algorithm of good wave filtering effect, strong robustness is target detection technique research.

Summary of the invention

The present invention be directed to the above-mentioned technical matters that existing small target infrared image disposal route exists, provide a kind of small target infrared image disposal route based on weighting topography entropy.

Based on a small target infrared image disposal route for weighting topography entropy, comprise the following steps:

Based on a small target infrared image disposal route for weighting topography entropy, comprise the following steps:

Step 1, solve the multiple dimensioned gray difference D of each pixel (x, y) of image;

Step 2, solve the topography entropy E of each pixel (x, y) of image;

Step 3, obtained the weighting topography entropy H of each pixel (x, y) by multiple dimensioned gray difference D and local image entropy E;

Step 4, solve adaptive threshold T according to weighting topography entropy H, and by adaptive threshold T, binaryzation is carried out to weighting topography entropy H, detect infrared small target.

The multiple dimensioned gray difference D of step 1 as above is solved by following steps:

Step 1.1, be I (x, y) for the gray-scale value that each pixel (x, y) in infrared image I is corresponding, the maximum neighborhood space Ω of pixel (x, y) is set _max, neighborhood space Ω _maxsize be L _max× L _max, wherein L _maxfor being greater than the positive odd number of 1;

Step 1.2, obtain the neighborhood space collection { Ω of each pixel (x, y) _k| k=1,2 ..., L}, wherein L=(L _max-1)/2, Ω _ksize be (2k+1) × (2k+1);

Step 1.3, utilize the neighborhood Ω of each pixel (x, y) of following formulae discovery _kwith Ω _maxbetween gray difference D _k(x, y), k=1,2 ..., L:

$D_k(x,y)={|\frac{1}{N_{{\mathrm{\Ω}}_k}}\underset{(s,t)\∈{\mathrm{\Ω}}_k}{\mathrm{\Σ}}I(s,t)-\frac{1}{N_{{\mathrm{\Ω}}_{\max }}}\underset{(p,q)\∈{\mathrm{\Ω}}_{\max }}{\mathrm{\Σ}}I(p,q)|}^2,k=\mathrm{1,2},...,L$

Wherein, with represent neighborhood Ω respectively _k, Ω _maxthe number of interior pixel, I (s, t) represents neighborhood Ω _kthe gray-scale value at interior point (s, t) place, I (p, q) represents neighborhood Ω _maxthe gray-scale value at interior point (p, q) place;

Step 1.4, calculate corresponding to each pixel (x, y) multiple dimensioned gray difference D (x, y):

D(x,y)＝max{D ₁(x,y),D ₂(x,y),...,D _L(x,y)}。

The topography entropy E of step 2 as above is solved by following steps:

The neighborhood space Θ of each pixel (x, y) in setting infrared image I, the size of neighborhood space Θ is m × n, calculates topography's entropy at pixel (x, y) place:

$E(x,y)=-\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}p\left(I(i,j)\right)\·{\log }_2(p\left(I(i,j)\right)+\mathrm{\ϵ}),p\left(I(i,j)\right)=\frac{I(i,j)}{{\mathrm{\Σ}}_{i=0}^{m-1}{\mathrm{\Σ}}_{j=0}^{n-1}I(i,j)}$

Wherein, ε is the normal number of setting, and I (i, j) represents the gray-scale value at point (i, the j) place in neighborhood Θ, and each pixel in traversal infrared image I, obtains the topography entropy E of infrared image I.

The weighting topography entropy H of step 3 as above is solved by following steps:

To each pixel (x, y) process the multiple dimensioned gray difference D that obtains through step 1 and process through step 2 the topography entropy E obtained and carry out dot-product operation, obtain the weighting topography entropy H that each pixel (x, y) is corresponding.

Adaptive threshold T as above is determined by following formula:

T＝c·SNR·σ+mm,SNR＝(H _max-mm)/σ

Wherein, c is positive constant, and σ is the standard deviation of weighting topography entropy H, and mm is the average of weighting topography entropy H, H _maxfor the maximal value of weighting topography entropy H.

The present invention compared with prior art, has the following advantages:

1. present invention utilizes the feature of target and background in small target infrared image, do not rely on infrared image model and parameter to select, can effectively suppress infrared image background and noise, improve the signal to noise ratio (S/N ratio) of infrared image, thus improve the detection probability of target, reduce false-alarm probability.

2. first the present invention builds the multiple dimensioned gray difference figure of infrared image, can reject a large amount of noise; Secondly obtain weighting topography entropy by dot-product operation, the weighting topography entropy diagram obtained has very high snr gain, can Background suppression and noise effectively; Then utilize adaptive threshold to detect target, avoid the problems such as the unstable and adaptivity of image procossing under complex background condition.

Accompanying drawing explanation

Fig. 1 is FB(flow block) of the present invention.

Fig. 2 is the comparison diagram of the result schematic diagram of result schematic diagram and the prior art algorithm adopting the present embodiment 1 method to obtain.A is the infrared original image of Small object of a width sea-empty background, and B is the filter result adopting multiple dimensioned gray difference operator, and C is the filter result adopting topography's Entropy algorithm, and D is weighting topography entropy diagram, and E is the testing result adopting adaptive threshold.

Fig. 3 is the infrared image processing result schematic diagram adopting prior art and the present embodiment method to obtain.(A_1), (B_1), (C_1), (D_1): be followed successively by the low signal-to-noise ratio small target infrared image under different background and noise level; (A_2), (B_2), (C_2), (D_2): be corresponding in turn in (A_1), (B_1), (C_1), the filter result based on maximum background forecast model method of (D_1); (A_3), (B_3), (C_3), (D_3): be corresponding in turn in (A_1), (B_1), (C_1), the filter result based on top cap operator of (D_1); (A_4), (B_4), (C_4), (D_4): be corresponding in turn in (A_1), (B_1), (C_1), the filter result of employing the present embodiment method step 1 ~ step 3 of (D_1); (A_5), (B_5), (C_5), (D_5): be corresponding in turn in (A_1), (B_1), (C_1), the infrared small target detection result based on the present embodiment method of (D_1).

Embodiment

Below by embodiment, and by reference to the accompanying drawings, technical scheme of the present invention is described in further detail.

Embodiment 1:

Fig. 1 is that this method mainly comprises the following steps: image inputs, and multiple dimensioned gray difference operator solves, and topography's Entropy algorithm solves, dot-product operation, and adaptive threshold solves, binaryzation.

Be specially:

Step 1, inputs a width infrared image, solves the multiple dimensioned gray difference D of image:

Small target infrared image generally by target, background and noise three part form.The imaging size of Small object is generally less than 80 pixels, and be namely less than 256 × 256 0.12%, thus target does not have size, shape and Texture eigenvalue, but it there are differences with background, noise in gray-scale value, frequency and correlativity etc.The core concept of multiple dimensioned gray difference operator (D) is the gray scale difference opposite sex utilized between target area in small target infrared image and target neighborhood, by the tolerance of otherness with Background suppression, strengthen target.

The solution procedure of the multiple dimensioned gray difference operator D of infrared image I is as follows:

(1) for each pixel (x, y) in infrared image I, corresponding gray-scale value is I (x, y), arranges the maximum neighborhood space Ω of pixel (x, y) _max, neighborhood space Ω _maxsize be L _max× L _max, wherein L _maxfor being greater than the positive odd number of 1;

(2) the neighborhood space collection { Ω of each pixel (x, y) is obtained _k| k=1,2 ..., L}, wherein L=(L _max-1)/2, Ω _ksize be (2k+1) × (2k+1);

(3) the neighborhood Ω of each pixel (x, y) is calculated _kwith Ω _maxbetween gray difference D _k(x, y), k=1,2 ..., L:

$D_k(x,y)={|\frac{1}{N_{{\mathrm{\Ω}}_k}}\underset{(s,t)\∈{\mathrm{\Ω}}_k}{\mathrm{\Σ}}I(s,t)-\frac{1}{N_{{\mathrm{\Ω}}_{\max }}}\underset{(p,q)\∈{\mathrm{\Ω}}_{\max }}{\mathrm{\Σ}}I(p,q)|}^2,k=\mathrm{1,2},...,L---\left(1\right)$

Wherein, with represent neighborhood Ω respectively _k, Ω _maxthe number of interior pixel, I (s, t) represents neighborhood Ω _kthe gray-scale value at interior point (s, t) place, I (p, q) represents neighborhood Ω _maxthe gray-scale value at interior point (p, q) place.

(4) calculate corresponding to each pixel (x, y) multiple dimensioned gray difference D (x, y):

D(x,y)＝max{D ₁(x,y),D ₂(x,y),...,D _L(x,y)} (2)

Each pixel in traversal infrared image I, obtains the multiple dimensioned gray difference D (as shown in the B of Fig. 2) of infrared image I.As can be seen from the B of Fig. 2, the background of infrared image I is inhibited, and target is strengthened well.

Step 2, solves the topography entropy E of image:

For the background of infrared image I, textural characteristics is determined, when there is target in image, the textural characteristics of image is destroyed, and Small object is less for the entropy contribution of entire image, but in local window, the appearance of Small object can cause the strong variations of Local textural feature, thus also can there is larger change in its local entropy.Utilizing the appearance of target can cause topography's entropy that this characteristic of larger change occurs can Background suppression, enhancing target.

For each pixel (x, y) in infrared image I, arrange its neighborhood space Θ, the size of neighborhood space Θ is m × n.Calculate topography's entropy at pixel (x, y) place:

$E(x,y)=-\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}p\left(I(i,j)\right)\·{\log }_2(p\left(I(i,j)\right)+\mathrm{\ϵ}),p\left(I(i,j)\right)=\frac{I(i,j)}{{\mathrm{\Σ}}_{i=0}^{m-1}{\mathrm{\Σ}}_{j=0}^{n-1}I(i,j)}---\left(3\right)$

Wherein, ε is default normal number, as ε=10 ^-6, I (i, j) represents the gray-scale value at point (i, the j) place in neighborhood Θ.

Each pixel in traversal infrared image I, obtains the topography entropy E (as shown in the C of Fig. 2) of infrared image I.There is homogeneous region in the A of Fig. 2, according to principle of maximum entropy, the entropy in this region is comparatively large, the white portion as shown in the C of Fig. 2, but the appearance of target causes the gray feature of image local area to change, and the change of this gray feature is still visible in the C of Fig. 2.

Step 3, solves the weighting topography entropy H of image:

The multiple dimensioned gray difference D (as shown in the B of Fig. 2) of infrared image I and local image entropy E (as shown in the C of Fig. 2) all can realize background suppress to infrared image and targets improvement.Merge D and E, the background of infrared image is suppressed further, and target is strengthened further.

To each pixel (x, processing the multiple dimensioned gray difference D that obtains through step 1 and processing through step 2 the topography entropy E obtained and carry out dot-product operation y), obtain each pixel (x, y) the weighting topography entropy H corresponding to, realization suppresses further the background of infrared image and target strengthens further, namely

$H=D\⊗E---\left(4\right)$

The weighting topography entropy H of infrared image I is as shown in the D of Fig. 2.As can be seen from the D of Fig. 2, the background of infrared image I is suppressed well, and target is also strengthened well.

Step 4, solves adaptive threshold T:

Adaptive threshold T is solved to processing through step 1, step 2 and step 3 the weighting topography entropy H obtained, and by adaptive threshold T, binaryzation is carried out to weighting topography entropy H, detect infrared small target (binaryzation result is as shown in the E of Fig. 2).The defining method of adaptive threshold T is

T＝c·SNR·σ+mm,SNR＝(H _max-mm)/σ (5)

Wherein, c is positive constant, and σ is the standard deviation of weighting topography entropy H, and mm is the average of weighting topography entropy H, H _maxfor the maximal value of weighting topography entropy H.

Adopt the result of different Infrared Image Processing Method as shown in Figure 3, as can be seen from Figure 3, the effect that the present embodiment method obtains is best, wherein, maximum background forecast model method comes from document (H.Deng and J.G.Liu, Infrared small target detection based on the self-information map, Infrared Physics & Technology, 2011, 54 (2): 100-107.), top cap Operator Method comes from document (X.Z.Bai and F.G.Zhou, Analysis of new top-hat transformation and the application for infrared dim small target detection, Pattern Recognition, 2010, 43 (6): 2145-2156.).

Signal to noise ratio (S/N ratio) (SNR, signal-to-noise ratio) is adopted to carry out the filter effect (expression formula of SNR is with reference to formula (5)) of the different Infrared Image Processing Method of objective evaluation.Concrete numerical value is in table 1.

Table 1 adopts the SNR of the filter effect of different Infrared Image Processing Method to compare.

Specific embodiment described herein is only to the explanation for example of the present invention's spirit.Those skilled in the art can make various amendment or supplement or adopt similar mode to substitute to described specific embodiment, but can't depart from spirit of the present invention or surmount the scope that appended claims defines.

Claims (5)

1., based on a small target infrared image disposal route for weighting topography entropy, it is characterized in that, comprise the following steps:

Step 1, solve the multiple dimensioned gray difference D of each pixel (x, y) of image;

Step 2, solve the topography entropy E of each pixel (x, y) of image;

Step 3, obtained the weighting topography entropy H of each pixel (x, y) by multiple dimensioned gray difference D and local image entropy E;

Step 4, solve adaptive threshold T according to weighting topography entropy H, and by adaptive threshold T, binaryzation is carried out to weighting topography entropy H, detect infrared small target.

2. a kind of small target infrared image disposal route based on weighting topography entropy according to claim 1, is characterized in that, the multiple dimensioned gray difference D of described step 1 is solved by following steps:

Step 1.1, be I (x, y) for the gray-scale value that each pixel (x, y) in infrared image I is corresponding, the maximum neighborhood space Ω of pixel (x, y) is set _max, neighborhood space Ω _maxsize be L _max× L _max, wherein L _maxfor being greater than the positive odd number of 1;

Step 1.2, obtain the neighborhood space collection { Ω of each pixel (x, y) _k| k=1,2 ..., L}, wherein L=(L _max-1)/2, Ω _ksize be (2k+1) × (2k+1);

Step 1.3, utilize the neighborhood Ω of each pixel (x, y) of following formulae discovery _kwith Ω _maxbetween gray difference D _k(x, y), k=1,2 ..., L:

$D_k(x,y)={|\frac{1}{N_{{\mathrm{\Ω}}_k}}\underset{(s,t){\∈\mathrm{\Ω}}_k}{\mathrm{\Σ}}I(s,t)-\frac{1}{N_{{\mathrm{\Ω}}_{\max }}}\underset{(p,q){\∈\mathrm{\Ω}}_{\max }}{\mathrm{\Σ}}I(p,q)|}^2,k=\mathrm{1,2},...,L$

Wherein, with represent neighborhood Ω respectively _k, Ω _maxthe number of interior pixel, I (s, t) represents neighborhood Ω _kthe gray-scale value at interior point (s, t) place, I (p, q) represents neighborhood Ω _maxthe gray-scale value at interior point (p, q) place;

Step 1.4, calculate corresponding to each pixel (x, y) multiple dimensioned gray difference D (x, y):

D(x,y)＝max{D ₁(x,y),D ₂(x,y),...,D _L(x,y)}。

3. a kind of small target infrared image disposal route based on weighting topography entropy according to claim 1, is characterized in that, the topography entropy E of described step 2 is solved by following steps:

The neighborhood space Θ of each pixel (x, y) in setting infrared image I, the size of neighborhood space Θ is m × n, calculates topography's entropy at pixel (x, y) place:

$E(x,y)=-\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}p\left(I(i,j)\right)\·{\log }_2(p\left(I(i,j)\right)+\mathrm{\ϵ}),p\left(I(i,j)\right)=\frac{I(i,j)}{{\mathrm{\Σ}}_{i=0}^{m-1}{\mathrm{\Σ}}_{j=0}^{n-1}I(i,j)}$

Wherein, ε is the normal number of setting, and I (i, j) represents the gray-scale value at point (i, the j) place in neighborhood Θ, and each pixel in traversal infrared image I, obtains the topography entropy E of infrared image I.

4. a kind of small target infrared image disposal route based on weighting topography entropy according to claim 1, is characterized in that, the weighting topography entropy H of described step 3 is solved by following steps:

To each pixel (x, y) process the multiple dimensioned gray difference D that obtains through step 1 and process through step 2 the topography entropy E obtained and carry out dot-product operation, obtain the weighting topography entropy H that each pixel (x, y) is corresponding.

5. a kind of small target infrared image disposal route based on weighting topography entropy according to claim 1, it is characterized in that, described adaptive threshold T is determined by following formula:

T＝c·SNR·σ+mm,SNR＝(H _max-mm)/σ

Wherein, c is positive constant, and σ is the standard deviation of weighting topography entropy H, and mm is the average of weighting topography entropy H, H _maxfor the maximal value of weighting topography entropy H.