CN103345730A - Infrared image processing method based on lateral inhibition network - Google Patents

Abstract

The invention relates to an infrared image processing method based on a lateral inhibition network. The method comprises the following steps that S1, partial entropy in each pixel point region of an image is calculated so that a partial entropy matrix of the image can be obtained; S2, according to the partial entropy matrix of the whole image, a maximum value of the matrix is calculated and then a measurement operator is constructed; S3, an inhibition coefficient matrix is solved; S4, according to the results obtained through the S2 and the S3, the infrared image is processed and lateral inhibition network output is obtained. According to the different characteristics of an image saltation zone (including edges and details) and a smoothness zone (including backgrounds and noise) in an empty region, the measuring operator is constructed through the image partial entropy. Therefore, the lateral inhibition network has the capacity of edge identification and is strong in noise immunity. After the lateral inhibition network is combined with a filtering mechanism, the image is enhanced and the lateral inhibition network has the noise inhibiting effect.

Description

Infrared Image Processing Method based on lateral inhibition network

Technical field

The present invention relates to infrared image and strengthen technical field, be specifically related to a kind of Infrared Image Processing Method based on lateral inhibition network.

Background technology

The vision lateral inhibition is that Hartline and colleague thereof find in the electric Physiological Experiment process in that the king crab compound eye is done the earliest.Years of researches are the result show, also there is the lateral inhibition function in the human visual system, and this comes from the inhibiting effect between the adjacent optic nerve unit, and at the pretreatment stage of visual signal, the lateral inhibition effect is considered to play crucial effects.

Infrared image is the infrared emanation energy imaging by the detector receiving target, therefore compare with visible images, characteristics such as infrared image has that contrast is low, edge fog and signal to noise ratio (S/N ratio) are low, and lateral inhibition network is actually a Hi-pass filter on the spatial frequency, and one of topmost function is exactly outstanding frame, enhanced contrast, therefore can come infrared image is handled with lateral inhibition, the image high-frequency information is enhanced, and low-frequency information is inhibited simultaneously, thereby improves contrast.But because infrared image inevitably can be introduced noise, noise and edge details belong to high frequency, lateral inhibition network is when strengthening edge details, make the noise of image also obtain amplification, thereby influenced visual effect, the subsequent treatment that also is unfavorable for image, therefore how utilizing lateral inhibition network to suppress noise when strengthening image is an emphasis of studying in the last few years.

Improvement algorithm to lateral inhibition network mainly can be divided into two classes at present: a class is based on improving one's methods of transform domain, the most frequently used transform domain is based on the method for frequency domain, specifically in frequency domain, distinguish image high-frequency information and low-frequency information by Fourier transform exactly, handle respectively then, the image that the spatial domain is enhanced is changed in last contravariant, and these class methods have treatment effect preferably, but calculated amount is bigger, can not satisfy real-time processing requirements, so practicality is not high.Another kind ofly be based on improving one's methods of spatial domain, concrete is exactly according to picture noise and the marginal information characteristics in the spatial domain, handle in conjunction with filtering mechanism, these class methods have advantages such as algorithm is simple, processing speed is fast, real-time, are present topmost methods.

Xie Xiaofang, people such as Mao Xiaobo are in " research of improved lateral inhibition network algorithm for image enhancement " literary composition, analyzed the difference of image local neighborhood gray-scale value and noise information, the method that a kind of auto adapted filtering mechanism combines with lateral inhibition network has been proposed, can come self-adaptation to adjust the output of wave filter according to the average of the local neighborhood of image and the variance of variance and noise, though this method has suppressed noise to a certain extent when strengthening image, but need know noise variance information in advance, and noise variance information is unknown often in the actual conditions, therefore has big limitation.

By above analysis as can be known, the image border details after existing lateral inhibition network is handled is enhanced, and contrast is improved, but noise information has also obtained amplification.The improvement algorithm calculated amount of some lateral inhibition networks is big, is unfavorable for real-time processing, and limitation is bigger, is difficult in the actual conditions to be applied.

Summary of the invention

In view of this, it is strong to the purpose of this invention is to provide a kind of edge recognition capability, and noise immunity is strong, when strengthening image, noise is also had inhibiting Infrared Image Processing Method based on lateral inhibition network,

The objective of the invention is to realize by such technical scheme,

A kind of Infrared Image Processing Method based on lateral inhibition network may further comprise the steps:

S1: obtain the local entropy in each neighborhood of pixel points of image, obtain the local entropy matrix of image, further obtain entire image local entropy matrix;

S2: according to entire image local entropy matrix, obtain the matrix maximal value, then structure tolerance operator;

S3: generate the lateral inhibition matrix of coefficients;

S4: infrared image is handled according to step S2 and S3 gained result.

Further, described step S1 specifically comprises following substep:

S11: (m gets the neighborhood of a N * N size centered by n), counts the number of times that each gray level occurs in the neighborhood, obtains the probability P that each gray level occurs with pixel in image space _j

In the formula: P _jFor gray level j at pixel (m, the n) probability that occurs in the neighborhood, m _jFor having the total pixel number of gray level j in the neighborhood;

S12: obtain local entropy H in this neighborhood of pixel points:

L represents that maximum gray scale is other in the formula;

S13: obtain the local entropy in each neighborhood of pixel points of image, obtain image local entropy matrix H.

Further, the value of the other L of described maximum gray scale is 256.

Further, described step S2 specifically comprises following substep:

S21: the maximal value H that obtains image local entropy matrix H _Max:

H _max＝max(H)

S22: (m is the gray average of neighborhood with radius l' size n) to obtain pixel

$\stackrel{\&OverBar;}{F(m,n)}=\frac{l^{\&prime;}}{{(2\&times;l^{\&prime;}+1)}^2}\underset{k1=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}\underset{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">k</figure-callout>}2=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}F(k1,k2),$

In the formula: k1, k2 represent pixel (m, n) each pixel coordinate in the radius l ' neighborhood;

S23: solve tolerance operator E, its computing formula is:

$E=\frac{H(m,n)}{H_{\max }}(F(m,n)-\stackrel{\&OverBar;}{F(m,n)})$

In the formula: H (m, n) the expression pixel (m, local entropy n), F (m, n) expression input image pixels point (m, gray-scale value n),

Expression pixel (m, n) neighborhood average.

Further, in described step S3, select bimodal Gauss model to solve the lateral inhibition matrix of coefficients, then the lateral inhibition coefficient table is shown: $k_{\mathrm{mn},\mathrm{pq}}=\frac{1}{\mathrm{\&beta;}}[\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_1}\exp (-\frac{d_{\mathrm{mn},\mathrm{<figure-callout\; id="2"\; label="pq"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">pq</figure-callout>}}^2}{{2\mathrm{\&sigma;}}_1^2})-\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_2}\exp (-\frac{d_{\mathrm{mn},\mathrm{<figure-callout\; id="2"\; label="pq"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">pq</figure-callout>}}^2}{{2\mathrm{\&sigma;}}_2^2})],$ β, σ in the formula ₁, σ ₂All be constant,

(m, n) (p, Euclidean distance q) are selected suitable neighborhood to the expression central pixel point, obtain the lateral inhibition matrix of coefficients with pixel.

Further, in described step S4, select acyclic network to calculate the output of lateral inhibition network, the output of lateral inhibition network is expressed as: $G(m,n)=\stackrel{\&OverBar;}{F(m,n)}-\underset{p=-l}{\overset{l}{\mathrm{\&Sigma;}}}\underset{q=-l}{\overset{l}{\mathrm{\&Sigma;}}}{k}_{\mathrm{mn},\mathrm{pq}}F(m+p,n+q)+\mathrm{\&lambda;E},$ K wherein _{Mn, pq}Be the lateral inhibition coefficient, λ is one and regulates constant that l is the lateral inhibition radius.

Further, described λ span is [0,1].

Owing to adopted technique scheme, the present invention to have following advantage:

The present invention proposes the Infrared Image Processing Method based on lateral inhibition network, in the lateral inhibition network input, use pixel (m, n) gray average in the neighborhood

(m n), makes lateral inhibition network have certain filter action to replace F.In lateral inhibition network, add tolerance operator E, can make network area tell edge, evenly and noise spot, these points are taked different inhibition strengths, and can make on the edge relative brightness of high brightness point bigger, the relative brightness of low-light level point is littler, the marginal point contrast is more obvious like this, has simultaneously also suppressed noise to a certain extent, and that can also avoid that filtering mechanism brings is fuzzy.Also may there be noise spot on the edge in addition, therefore when strengthening the edge, also strengthened the noise on the edge, but human eye will be far smaller than level and smooth district for noise-sensitive degree on the edge, and the tolerance operator of the present invention's structure meets human-eye visual characteristic from this point.

Other advantages of the present invention, target and feature will be set forth to a certain extent in the following description, and to a certain extent, based on being apparent to those skilled in the art to investigating hereinafter, perhaps can obtain instruction from the practice of the present invention.Target of the present invention and other advantages can realize and obtain by following instructions.

Description of drawings

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is described in further detail below in conjunction with accompanying drawing, wherein:

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is original image;

Fig. 3 is the local entropy image corresponding with Fig. 2;

Fig. 4 is the local entropy value that black line is capable among Fig. 2;

Fig. 5 strengthens image for lateral inhibition network;

Fig. 6 is that improved lateral inhibition network strengthens image.

Embodiment

Below with reference to accompanying drawing, the preferred embodiments of the present invention are described in detail; Should be appreciated that preferred embodiment only for the present invention is described, rather than in order to limit protection scope of the present invention.

As shown in Figure 1, a kind of Infrared Image Processing Method based on lateral inhibition network may further comprise the steps: the local entropy of image is found the solution; Structure tolerance operator; Find the solution the lateral inhibition matrix of coefficients; Lateral inhibition network is used for image to be handled.

Entropy is mainly used in representing the uncertainty of things state, and image entropy has reflected that what of quantity of information image contain, and specifically is exactly to have represented the space characteristics that gradation of image distributes.The neighborhood territory pixel gray level is more in the edge details district, and then local entropy is bigger, illustrates that to contain quantity of information many, and background area neighborhood gray level is less, and local entropy is less, and it is few to contain quantity of information.If (m n) gets the neighborhood of a N * N size, and the entropy of trying to achieve so is exactly pixel (m, local entropy n) at the aerial image vegetarian refreshments of image.

S1: obtain the local entropy in each neighborhood of pixel points of image, obtain the local entropy matrix of image, further obtain entire image local entropy matrix;

S11: (m gets the neighborhood of a N * N size centered by n), counts the number of times that each gray level occurs in the neighborhood, obtains the probability P that each gray level occurs with pixel in image space _j In the formula: P _jFor gray level j at pixel (m, the n) probability that occurs in the neighborhood, m _jFor having the total pixel number of gray level j in the neighborhood;

S12: obtain local entropy H in this neighborhood of pixel points:

$H=-\underset{j=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{P}_j\log P_j$

L represents that maximum gray scale is other in the formula; The value of general L is 256;

S13: travel through this image, obtain the local entropy in each neighborhood of pixel points of image, obtain image local entropy matrix H.

Local entropy is only relevant with image local grey scale change degree, have nothing to do and change size with local gray-value, and the infrared image edge gray-value variation is often very little, and the edge is littler a little less than some, therefore is distributed with its special advantages with local entropy reflection infrared image local gray level.

S2: according to entire image local entropy matrix, obtain the matrix maximal value, then structure tolerance operator;

S21: the maximal value H that obtains image local entropy matrix H _Max:

H _max＝max(H)

S22: (m is the gray average of neighborhood with radius l' size n) to obtain pixel

$\stackrel{\&OverBar;}{F(m,n)}=\frac{l^{\&prime;}}{{(2\&times;l^{\&prime;}+1)}^2}\underset{k1=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}\underset{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">k</figure-callout>}2=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}F(k1,k2),$

S23: solve tolerance operator E, its computing formula is:

$E=\frac{H(m,n)}{H_{\max }}(F(m,n)-\stackrel{\&OverBar;}{F(m,n)})$

In the formula: H (m, n) the expression pixel (m, local entropy n), F (m, n) expression input image pixels point (m, gray-scale value n),

Expression pixel (m, n) neighborhood average.

For marginal point, H (m, n) very big,

Absolute value also bigger, so the E absolute value is very big in the equation.For level and smooth district noise spot H (m, n) less,

Bigger, so E is bigger.For the even point in level and smooth district, H (m, n) less,

Less, so the E minimum.(m n) is the high brightness point, then as if point

E be bigger on the occasion of; If (m n) is the low-light level point, then to point E is bigger negative value.So the tolerance operator has just been distinguished edge, noise and background.

S3: generate the lateral inhibition matrix of coefficients;

Experiment shows that the inhibiting effect of receptor unit reduced along with the increase of distance between the receptor unit around arbitrary receptor unit was suffered, so the lateral inhibition coefficient can be regarded the function apart from d as.Lateral inhibition coefficient model mainly contains at present: hyperbolic model, Gauss model, bimodal Gauss model, exponential model.The trend of monotone decreasing in these models all satisfy between designation area has specifically just reflected that the distance between inhibiting intensity and the receptor unit is inversely proportional in lateral inhibition, namely distance is more big, and inhibition strength is more little.

Studies show that had the receptor of a segment distance also to be eager to excel than the most contiguous ommatidium effect from trying receptor, therefore bimodal Gauss model is near the vision actual conditions, and then the lateral inhibition coefficient table is shown:

$k_{\mathrm{mn},\mathrm{pq}}=\frac{1}{\mathrm{\&beta;}}[\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_1}\exp (-\frac{d_{\mathrm{mn},\mathrm{<figure-callout\; id="2"\; label="pq"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">pq</figure-callout>}}^2}{{2\mathrm{\&sigma;}}_1^2})-\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_2}\exp (-\frac{d_{\mathrm{mn},\mathrm{<figure-callout\; id="2"\; label="pq"\; filenames="HDA00003526109500014.png"\; state="\{\{state\}\}">pq</figure-callout>}}^2}{{2\mathrm{\&sigma;}}_2^2})]$

β, σ in the formula ₁, σ ₂All be constant,

(m, n) (p, Euclidean distance q) are selected suitable neighborhood to the expression central pixel point, obtain the lateral inhibition matrix of coefficients with pixel.

S4: infrared image is handled according to step S2 and S3 gained result.

Lateral inhibition mechanism is because the interaction between the visual receptor unit produces, it is a kind of mode of vision system process information, on the one hand space contrast information is enhanced, the identical information in space is inhibited, thereby strengthened edge details, made the clear of fuzzy image change.The lateral inhibition network model has many, wherein usefulness is the Hartline-Ratliff model the most widely, be divided into circulation lateral inhibition network and acyclic lateral inhibition network, wherein circulation side inhibition network calculations amount is very big, be difficult to accomplish the real-time processing of image, so we selects acyclic network for use, and operator is measured in adding in network, with rejection coefficient matrix and each pixel convolution of image of trying to achieve, carry out infrared image processing then, the output of the lateral inhibition network that obtains at last can be expressed as:

$G(m,n)=\stackrel{\&OverBar;}{F(m,n)}-\underset{p=-l}{\overset{l}{\mathrm{\&Sigma;}}}\underset{q=-l}{\overset{l}{\mathrm{\&Sigma;}}}{k}_{\mathrm{mn},\mathrm{pq}}F(m+p,n+q)+\mathrm{\&lambda;E}$

K wherein _{Mn, pq}Be the lateral inhibition coefficient, λ is one and regulates constant that general λ span is [0,1], increased the dirigibility of lateral inhibition network processing image like this, l is the lateral inhibition radius, in order to make treatment effect better, average radius l ' can get different values with lateral inhibition radius l.

Fig. 2 represents original image, and Fig. 3 is the local entropy image (black line among the figure is capable for the artificial sign that adds) of the former figure that obtains.As seen from the figure, the local entropy of image border point is greater than background area, and Fig. 4 has provided and drawn the capable local entropy value of black line among Fig. 3, as can be seen, though the noise of background area has caused influence to local entropy, but still less than the marginarium.

Be the image that former lateral inhibition network strengthens by Fig. 5, when image was enhanced as can be seen, noise had also obtained amplification, and visual effect is bad.The image that Fig. 6 strengthens for the present invention, noise has also obtained inhibition in the time of the figure image intensifying, and is better than former lateral inhibition network treatment effect.

The present invention has analyzed the deficiency of existing lateral inhibition network, has proposed a kind of lateral inhibition network and has improved one's methods, and uses pixel (m, n) gray average in the neighborhood in the lateral inhibition network input

(m n), makes lateral inhibition network have certain filter action to replace F.In lateral inhibition network, add tolerance operator E, can make network area tell edge, evenly and noise spot, these points are taked different inhibition strengths, and can make on the edge relative brightness of high brightness point bigger, the relative brightness of low-light level point is littler, the marginal point contrast is more obvious like this, has simultaneously also suppressed noise to a certain extent, and that can also avoid that filtering mechanism brings is fuzzy.The present invention improves one's methods for a kind of lateral inhibition network, and infrared image enhancing technology is had theory and practical value preferably.

The above is the preferred embodiments of the present invention only, is not limited to the present invention, and obviously, those skilled in the art can carry out various changes and modification and not break away from the spirit and scope of the present invention the present invention.Like this, if of the present invention these are revised and modification belongs within the scope of claim of the present invention and equivalent technologies thereof, then the present invention also is intended to comprise these changes and modification interior.

Claims (7)

1. Infrared Image Processing Method based on lateral inhibition network is characterized in that: may further comprise the steps:

S1: obtain the local entropy in each neighborhood of pixel points of image, obtain the local entropy matrix of image, further obtain entire image local entropy matrix;

S2: according to entire image local entropy matrix, obtain the matrix maximal value, then structure tolerance operator;

S3: generate the lateral inhibition matrix of coefficients;

S4: infrared image is handled according to step S2 and S3 gained result.

2. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 1, it is characterized in that: described step S1 specifically comprises following substep:

S11: (m gets the neighborhood of a N * N size centered by n), counts the number of times that each gray level occurs in the neighborhood, obtains the probability P that each gray level occurs with pixel in image space _j

In the formula: P _jFor gray level j at pixel (m, the n) probability that occurs in the neighborhood, m _jFor having the total pixel number of gray level j in the neighborhood;

S12: obtain local entropy H in this neighborhood of pixel points:

L represents that maximum gray scale is other in the formula;

S13: obtain the local entropy in each neighborhood of pixel points of image, obtain image local entropy matrix H.

3. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 2, it is characterized in that: the value of the other L of described maximum gray scale is 256.

4. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 1, it is characterized in that: described step S2 specifically comprises following substep:

S21: the maximal value H that obtains image local entropy matrix H _Max:

H _max＝max(H)

S22: (m is the gray average of neighborhood with radius l' size n) to obtain pixel

$\stackrel{\&OverBar;}{F(m,n)}=\frac{l^{\&prime;}}{{(2\&times;l^{\&prime;}+1)}^2}\underset{k1=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}\underset{k2=-l^{\&prime;}}{\overset{l^{\&prime;}}{\mathrm{\&Sigma;}}}F(k1,k2),$

In the formula: k1, k2 represent pixel (m, n) each pixel coordinate in the radius l ' neighborhood;

S23: solve tolerance operator E, its computing formula is:

$E=\frac{H(m,n)}{H_{\max }}(F(m,n)-\stackrel{\&OverBar;}{F(m,n)})$

In the formula: H (m, n) the expression pixel (m, local entropy n), F (m, n) expression input image pixels point (m, gray-scale value n),

Expression pixel (m, n) neighborhood average.

5. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 1, it is characterized in that: in described step S3, select bimodal Gauss model to solve the lateral inhibition matrix of coefficients, then the lateral inhibition coefficient table is shown:

$k_{\mathrm{mn},\mathrm{pq}}=\frac{1}{\mathrm{\&beta;}}[\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_1}\exp (-\frac{d_{\mathrm{mn},\mathrm{pq}}^2}{{2\mathrm{\&sigma;}}_1^2})-\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_2}\exp (-\frac{d_{\mathrm{mn},\mathrm{pq}}^2}{{2\mathrm{\&sigma;}}_2^2})],$

β, σ in the formula ₁, σ ₂All be constant,

(m, n) (p, Euclidean distance q) are selected suitable neighborhood to the expression central pixel point, obtain the lateral inhibition matrix of coefficients with pixel.

6. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 1 is characterized in that: in described step S4, select acyclic network to calculate the output of lateral inhibition network, then the output of lateral inhibition network is expressed as: $G(m,n)=\stackrel{\&OverBar;}{F(m,n)}-\underset{p=-l}{\overset{l}{\mathrm{\&Sigma;}}}\underset{q=-l}{\overset{l}{\mathrm{\&Sigma;}}}{k}_{\mathrm{mn},\mathrm{pq}}F(m+p,n+q)+\mathrm{\&lambda;E},$ K wherein _{Mn, pq}Be the lateral inhibition coefficient, λ is one and regulates constant that l is the lateral inhibition radius.

7. a kind of Infrared Image Processing Method based on lateral inhibition network according to claim 6 is characterized in that: described λ span is [0,1].

Images