CN103606130A - Infrared degraded image adaptive restoration method - Google Patents

Abstract

The invention relates to an infrared degraded image adaptive restoration method. The infrared degraded image adaptive restoration method comprises the following steps that: (1) an input image I [in] is obtained; (2) an image degradation model A is constructed; (3) the information entropy of the image I [in] is calculated; (4) a regularization coefficient Lambda is calculated according to H; (5) an image I can be restored through iterative operation; and (6) an restored image I [out] is outputted. Compared with the prior art, the infrared degraded image adaptive restoration method of the invention has the following advantages that: (1) the infrared degraded image adaptive restoration method is especially designed for low signal-to-noise-ratio infrared images, and detail information can be effectively restored when noise is suppressed; (2) the regularization parameter can be calculated adaptively according to the information entropy of local regions of the image, and the infrared degraded image adaptive restoration method has a multi-scale restoring ability, and can realize a strong regularization ability for image smooth regions and a strong regularization ability for image detail regions; (3) the image degradation model can be flexibly constructed according to an actual degradation process; and (4) higher-order computation and complex structure do not exist, and the quantity of computation is small, and therefore, hardware real-time implementation is easy.

Description

Infrared degraded image adaptive restoration method

Technical field

The present invention designs a kind of infrared degraded image adaptive restoration method, particularly a kind of image recovery method of applicable hardware real-time implementation.

Background technology

In infrared image acquisition, transmission and processing procedure, owing to affected by the factors such as relative motion of atmospheric disturbance, optical system poor focusing, scenery and imaging device, cause the image quality decrease that obtains, image thickens.In order to obtain the image of high s/n ratio, high definition, need to degraded image, restore according to image degradation model.

Image restoration is the processing procedure of basic a, prerequisite in image processing field, in primary vision is processed, occupies extremely important status, and Chinese scholars all takes much count of the research of this respect.How much image restoration can be divided into typical image restoration and the large class of blind image restoration two according to known priori.The complexity of image restoration depends primarily on the levels of precision that priori is grasped.Typical image restoration is the point spread function of degenerating according to definite priori computing system, then utilize the anti-method of degenerating such as liftering, least squares filtering, degraded image is restored, as liftering method, airspace filter method and algebraic method etc.Blind image restoration refers to the priori that does not need the priori of system degradation or only need part system degradation, by degenerative process (fuzzy and noise) is set up to model, and then estimate true picture from degraded image feature, as zero blade face partition method, ARMA Parameter Estimation Method, priori fuzzy recognition method etc.There is following shortcoming in existing infrared image restored method: (1) most existing infrared image restored methods are only applicable to the input picture of high s/n ratio, along with input picture signal to noise ratio (S/N ratio) declines, restored image visual effect variation, unfavorable with the mankind or machine to the understanding of image and analysis; (2) most existing infrared image restored methods are used same regularization parameter to carry out regularization to entire image, do not utilize local feature information, cause having lost in recuperation a large amount of detailed information; (3) most existing image recovery method operands are large, are not easy to hardware real-time implementation.

Summary of the invention

Goal of the invention: the object of the invention is to design that a kind of method is simple, applicability is strong, the infrared degraded image adaptive restoration method of respond well and applicable hardware real-time implementation.

Technical scheme: the implementation step of technical solution of the present invention is as follows: infrared degraded image adaptive restoration method, comprises the steps:

(1) obtain input picture I _in;

(2) construct image degradation model A;

(3) calculate I _ininformation entropy H;

(4) according to H, calculate regularization coefficient lambda;

(5) by interative computation, image I is restored;

(6) output restored image I _out.

In the infrared degraded image adaptive restoration of the present invention method, image degradation model A selects gauss low frequency filter, and two-dimentional expression formula is as follows:

$A(u,v)=e^{-\frac{D^2(u,v)}{2{\mathrm{\σ}}^2}}$

Wherein, A (u, v) represents that degradation model A is at the coefficient of coordinate (u, v) position, and D (u, v) is the distance apart from Fourier transform initial point, and σ represents the degree of Gaussian curve expansion, and u and v represent respectively horizontal ordinate and ordinate.

In the infrared degraded image adaptive restoration of the present invention method, the information entropy H expression formula of the P * Q neighborhood centered by coordinate (i, j) is as follows:

$H(i,j)=-\underset{m=1}{\overset{P}{\mathrm{\Σ}}}\underset{n=1}{\overset{Q}{\mathrm{\Σ}}}p(i,j)\·\mathrm{lgp}(i,j)$

$p(i,j)=I_{\mathrm{in}}(i,j)/\left[\underset{m=1}{\overset{P}{\mathrm{\Σ}}}\underset{n=1}{\overset{Q}{\mathrm{\Σ}}}{I}_{\mathrm{in}}(i+m,j+n)\right]$

Wherein, H (i, j) represents the information entropy of H coordinate (i, j) position, the intensity profile probability of p (i, j) denotation coordination (i, j) position, I _in(i, j) and I _in(i+m, j+n) be presentation video I respectively _inthe gray-scale value of coordinate (i, j) and (i+m, j+n) position, 1≤m≤P, 1≤n≤Q.

In the infrared degraded image adaptive restoration of the present invention method, regularization coefficient lambda expression formula is as follows:

$\mathrm{\λ}(i,j)=\frac{1}{\frac{H(i,j)-H_{\min }}{H_{\max }-H_{\min }}+\mathrm{\β}}$

Wherein, λ (i, j) represents the regularization coefficient of λ coordinate (i, j) position, H _maxand H _minrepresent respectively maximal value and minimum value in information entropy H, β is constant coefficient, and avoiding denominator is 0.

In the infrared degraded image adaptive restoration of the present invention method, the k+1 time iterative image I ^k+1expression formula is as follows:

$I^{k+1}(i,j)=I^k(i,j)\left[\right[\left[\frac{I_{\mathrm{in}}}{(I^k*A)(i,j)}\right]\⊕A](i,j)-\mathrm{\λ}(i,j)\·U(i,j)]$

Wherein, I ^k+1(i, j) presentation video I ^k+1the gray-scale value of coordinate (i, j) position, I ^k(i, j) presentation video I ^kthe gray-scale value of coordinate (i, j) position, U (i, j) represents the regularization value of regularization factor U coordinate (i, j) position, A presentation video degradation model, * is convolution algorithm, ⊕ is related operation, is phase multiplication.

In the infrared degraded image adaptive restoration of the present invention method, restored image I _outexpression formula is as follows:

I _out=I ^k+1

Wherein, k represents iterations.

Beneficial effect: the present invention compared with prior art, has following remarkable advantage: (1), especially for Low SNR Infrared Images, can have efficient recovery detailed information when suppressing noise; (2) according to the information entropy self-adaptation of image local area, calculate regularization parameter, possess multiple dimensioned restorability, realized image smoothing district regularization ability strong, a little less than image detail region regularization ability; (3) image degradation model can build flexibly according to actual degenerative process; (4) do not have high exponent arithmetic(al) and labyrinth, algorithm operation quantity is little, is easy to hardware real-time implementation.

Accompanying drawing explanation

Accompanying drawing is the process flow diagram of the infrared degraded image adaptive restoration of the present invention method.

Embodiment

By reference to the accompanying drawings, with example, describe technical scheme of the present invention in detail below:

The present invention relates to infrared degraded image adaptive restoration method, thermal infrared imager focal plane arrays (FPA) size is 320 * 256, and working frame frequency is 25 frames per second.Image processing platform adopts DSP+FPGA framework, and infrared degraded image adaptive restoration method realizes in dsp processor, meets the demand of processing in real time, and concrete implementation step is as follows:

(1) obtain input picture I _in;

Dsp processor input picture I _inbe 14 bit digital images, picture size is 320 * 256.

(2) construct image degradation model A;

Image degradation model is selected gauss low frequency filter, and it is 5 * 5 that filter size is set, and standard deviation is 0.6, and electric-wave filter matrix is expressed as shown in table 1:

Table 15 * 5 gauss low frequency filters

0	0.0004	0.0017	0.0004	0
					0.0004	0.0274	0.1099	0.0274	0.0004
0.0017	0.1099	0.4407	0.1099	0.0017
					0.0004	0.0274	0.1099	0.0274	0.0004
0	0.0004	0.0017	0.0004	0

(3) calculate I _ininformation entropy H;

Select 8 * 8 neighborhood computing information entropys, expression formula is as follows:

$H(i,j)=-\underset{m=1}{\overset{8}{\mathrm{\Σ}}}\underset{n=1}{\overset{8}{\mathrm{\Σ}}}p(i,j)\·\mathrm{lgp}(i,j)$

$p(i,j)=I_{\mathrm{in}}(i,j)/\left[\underset{m=1}{\overset{8}{\mathrm{\Σ}}}\underset{n=1}{\overset{8}{\mathrm{\Σ}}}{I}_{\mathrm{in}}(i+m,j+n)\right]$

Wherein, H (i, j) represents the information entropy of H coordinate (i, j) position, the intensity profile probability of p (i, j) denotation coordination (i, j) position, I _in(i, j) and I _in(i+m, j+n) be presentation video I respectively _inthe gray-scale value of coordinate (i, j) and (i+m, j+n) position, 1≤m≤8,1≤n≤8,1≤i≤256,1≤j≤320.

(4) according to H, calculate regularization coefficient lambda;

By comparison operation, try to achieve the maximal value H in H _maxwith minimum value H _min, regularization coefficient lambda expression formula is as follows:

$\mathrm{\λ}(i,j)=\frac{1}{\frac{H(i,j)-H_{\min }}{H_{\max }-H_{\min }}+\mathrm{\β}}$

Wherein, λ (i, j) represents the regularization coefficient of λ coordinate (i, j) position, β=0.001,1≤i≤256,1≤j≤320.

(5) by interative computation, image I is restored;

The k+1 time iterative image I ^k+1expression formula is as follows:

$I^{k+1}(i,j)=I^k(i,j)\left[\right[\left[\frac{I_{\mathrm{in}}}{(I^k*A)(i,j)}\right]\⊕A](i,j)-\mathrm{\λ}(i,j)\·U(i,j)]$

Wherein, I ^k+1(i, j) presentation video I ^k+1the gray-scale value of coordinate (i, j) position, I ^k(i, j) presentation video I ^kthe gray-scale value of coordinate (i, j) position, U (i, j) represent regularization factor U coordinate (i, j) the regularization value of position, selects Tikhonov model to calculate, A presentation video degradation model, * be convolution algorithm, ⊕ is related operation, is phase multiplication, k=10,1≤i≤256,1≤j≤320.

(6) output restored image I _out.

I _out=I ^k+1

Wherein, k=10.

Claims (6)

1. infrared degraded image adaptive restoration method, is characterized in that: comprise the steps:

(1) obtain input picture I _in;

(2) construct image degradation model A;

(3) calculate I _ininformation entropy H;

(4) according to H, calculate regularization coefficient lambda;

Regularization coefficient lambda expression formula is as follows:

$\mathrm{\λ}(i,j)=\frac{1}{\frac{H(i,j)-H_{\min }}{H_{\max }-H_{\min }}+\mathrm{\β}}$

Wherein, λ (i, j) represents the regularization coefficient of λ coordinate (i, j) position, H _maxand H _minrepresent respectively maximal value and minimum value in information entropy H, β is constant coefficient, and avoiding denominator is 0;

(5) by interative computation, image I is restored;

(6) output restored image I _out.

2. infrared degraded image adaptive restoration method according to claim 1, is characterized in that, in described step (2), image degradation model A selects gauss low frequency filter, and two-dimentional expression formula is as follows:

$A(u,v)=e^{-\frac{D^2(u,v)}{2{\mathrm{\σ}}^2}}$

Wherein, A (u, v) represents that degradation model A is at the coefficient of coordinate (u, v) position, and D (u, v) is the distance apart from Fourier transform initial point, and σ represents the degree of Gaussian curve expansion, and u and v represent respectively horizontal ordinate and ordinate.

3. infrared degraded image adaptive restoration method according to claim 1, is characterized in that, in described step (3), the information entropy H expression formula of the P * Q neighborhood centered by coordinate (i, j) is as follows:

$H(i,j)=-\underset{m=1}{\overset{P}{\mathrm{\Σ}}}\underset{n=1}{\overset{Q}{\mathrm{\Σ}}}p(i,j)\·\mathrm{lgp}(i,j)$

$p(i,j)=I_{\mathrm{in}}(i,j)/\left[\underset{m=1}{\overset{P}{\mathrm{\Σ}}}\underset{n=1}{\overset{Q}{\mathrm{\Σ}}}{I}_{\mathrm{in}}(i+m,j+n)\right]$

Wherein, H (i, j) represents the information entropy of H coordinate (i, j) position, the intensity profile probability of p (i, j) denotation coordination (i, j) position, I _in(i, j) and I _in(i+m, j+n) be presentation video I respectively _inthe gray-scale value of coordinate (i, j) and (i+m, j+n) position, 1≤m≤P, 1≤n≤Q.

4. infrared degraded image adaptive restoration method according to claim 1, is characterized in that, in described step (4), according to the information entropy self-adaptation of Image neighborhood, calculates regularization coefficient lambda.

5. infrared degraded image adaptive restoration method according to claim 1, is characterized in that, in described step (5), and the k+1 time iterative image I ^k+1expression formula is as follows:

$I^{k+1}(i,j)=I^k(i,j)\left[\right[\left[\frac{I_{\mathrm{in}}}{(I^k*A)(i,j)}\right]\⊕A](i,j)-\mathrm{\λ}(i,j)\·U(i,j)]$

Wherein, I ^k+1(i, j) presentation video I ^k+1the gray-scale value of coordinate (i, j) position, I ^k(i, j) presentation video I ^kthe gray-scale value of coordinate (i, j) position, U (i, j) represents the regularization value of regularization factor U coordinate (i, j) position, A presentation video degradation model, * is convolution algorithm, ⊕ is related operation, is phase multiplication.

6. infrared degraded image adaptive restoration method according to claim 1, is characterized in that, in described step (6), and restored image I _outexpression formula is as follows:

I _out=I ^k+1

Wherein, k represents iterations.

Images