CN103440630B - Show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter - Google Patents

Abstract

The invention discloses a kind of Larger Dynamic scope infrared image based on guiding wave filter to show and detail enhancing method.First the method applies the disposal of gentle filter guiding wave filter that original Larger Dynamic scope infrared image carries out edge reservation, obtain the Primary layer image information that details is fuzzy, this Primary layer image and original image are done difference, obtain image detail layer information, then levels of detail image being carried out self adaptation enhancement process, Primary layer image is carried out dynamic range compression process, the image after finally both being processed merges, and reject overflow value, obtain the low dynamic range echograms eventually for display.The method while compressing image data, can be effectively improved picture contrast, strengthens image detail, improves the visual effect of image；Meanwhile, the method amount of calculation is less, it is simple to practical application.

Description

Show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter

Technical field

The inventive method belongs to infrared image processing technical field, and particularly a kind of Larger Dynamic scope infrared image processed based on guiding filter filtering shows and detail enhancing method.

Background technology

Modern high performance thermal infrared imager is obtained in that the raw image data that dynamic range is very big, and bit wide is typically in 12-14 position, but the dynamic range bit wide of display device is generally 8.Generally when after the original image of acquisition Larger Dynamic scope, it is necessary to original image is remapped, by its dynamic range compression so that the display of display device.This process typically requires and reaches two conditions: the first, compression raw image data, enables the dynamic range of high-performance thermal infrared imager output image to match with the dynamic range of display device；The second, simultaneously, in compression process, retain the details existed in original image as far as possible, make observer finally be able to observe that the image with better visual effect on the display device, and can be easier to distinguish the weak target hidden in the background.

Conventional dynamic range compression algorithm, has automatic growth control and histogram equalization, is most popular image display technology in infrared imaging system.First automatic growth control rejects the extremum in raw image data, and then by overall dynamic range Linear Mapping to 8, but this method there will be, and contrast is low, the obvious problem of loss in detail.Histogram equalization output image then there will be the problems such as enhancing, the amplification of homogeneous area noise, bleaching effect.

Conventional detail enhancement algorithms, as contrast enhancement algorithms in many documents extensively visible (referring to document one: K.Zuiderveld, " Contrastlimitedadaptivehistogramequalizaiton; " inGraphicsGemsIV, pp.474-485, AcademicPressProfessional, Inc., SanDiego1994.).But contrast enhancement algorithms is mostly for low-dynamic range infrared image, Larger Dynamic scope original image is inapplicable；And only only account for the enhancing problem of rest image, it does not have consider the application in real-time system.

Considering the deficiency of automatic growth control and histogram equalizing method, many more complicated methods are suggested, such as Retinex, but these methods are primarily directed to visible images, namely for visible images, there is good effect, and to infrared image poor effect, there is the problem that noise amplifies；And for example based on the algorithm of layered shaping framework (referring to document two: ChaoZuo, QianChen, NingLiuetal.Displayanddetailenhancementforhigh-dynamic-r angeinfraredimages [J] .OpticalEngineering, 2011,50 (12): 127401 (9) .), its main thought is to divide the image into levels of detail and Primary layer, then it is acted upon respectively, but the method operand is big, cause real-time poor, and be easily generated gradient flop phenomenon.

Summary of the invention

The present invention proposes a kind of Larger Dynamic scope infrared image based on guiding wave filter and shows and detail enhancing method, make the picture contrast after process and image detail all been significantly enhanced, and operand is little, real-time, good without gradient flop phenomenon, noise suppression effect.

In order to solve above-mentioned technical problem, the present invention proposes a kind of Larger Dynamic scope infrared image based on guiding wave filter and shows and detail enhancing method, it is characterised in that comprise the following steps:

Step one: use and guide wave filter that original infrared picture data is carried out the disposal of gentle filter, it is thus achieved that the base layer data of original infrared image, and base layer data and original infrared picture data are done difference, it is thus achieved that levels of detail data；

Step 2: use and guide filtering core that levels of detail data carry out self adaptation enhancing calculating, it is thus achieved that the enhanced levels of detail data of self adaptation；

Step 3: use histogram projection technology that base layer data carries out dynamic range compression and contrast enhancement processing, it is thus achieved that the base layer data after dynamic range compression；

Step 4: the base layer data after dynamic range compression described in enhanced for self adaptation described in step 2 levels of detail data and step 3 is merged, and rejects overflow value, it is thus achieved that finally export view data；

Described step one use shown in the computational methods such as formula (1) guiding wave filter to calculate base layer data,

$I_B(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}\left(a\right(i^{\′},j^{\′}\left)I_{in}\right(i^{\′},j^{\′})+b(i^{\′},j^{\′}\left)\right)---\left(1\right)$

In formula (1), I_inFor the original infrared picture data of input, I_BFor base layer data, (i, j) for pixel coordinate, w_i,jIt is that (w is window w for i, the window of pixel centered by j) with pixel_i,jThe number of middle pixel, a_(i',j')And b_(i',j')It is the linear coefficient centered by (i', j') in the window of pixel, a_(i',j')Shown in computational methods such as formula (2), b_(i',j')Shown in computational methods such as formula (3),

$a(i,j)=\frac{\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}{I}_{in}(i^{\′},j^{\′})G(i^{\′},j^{\′})-{\mathrm{\μ}}_{i,j}\stackrel{\‾}{I_{in}(i^{\′},j^{\′})}}{{\mathrm{\σ}}_{i,j}+\mathrm{\ϵ}}.---\left(2\right)$

$b(i,j)=\stackrel{\‾}{I_{in}(i^{\′},j^{\′})}-a(i,j){\mathrm{\μ}}_{i,j}.---\left(3\right)$

In formula (2) and formula (3), μ_i',j'Represent the average of each pixel in window；σ_i',j'Represent the variance of each pixel in window；ε is linear regression coeffficient, for determining the smoothness of this wave filter；Represent given window w_i,jMiddle I_inAverage；G (i', j') is the navigational figure data guiding wave filter for processing, G (i', j')=I_in(i, j), namely raw image data itself is as navigational figure data；

In described step one, base layer data and original infrared picture data are done shown in the computational methods such as formula (4) of difference acquisition levels of detail data,

I_D(i, j)=I_in(i,j)-I_B(i,j)(4)

In formula (4), I_DRepresent image detail layer data；

In described step 2, self adaptation strengthens shown in computational methods such as formula (5),

I_DP=I_D*(W(i,j)*a+b)(5)

In formula (5), I_DPThe levels of detail data obtained after calculating after strengthening for self adaptation, a and b is Serial regulation coefficient, W (i, is j) guide filter kernel coefficient, W (i, shown in the such as formula of computational methods j) (6),

$W(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}\left(\frac{{\left(I\right(i,j)-{\mathrm{\μ}}_{i^{\′},j^{\′}})}^2}{{{\mathrm{\σ}}_{j^{\′},j^{\′}}}^2+\mathrm{\δ}}\right)---\left(6\right)$

In formula (6), (i is j) with pixel (i, the pixel value of the window of pixel each pixel interior, μ centered by j) to I_i',j'And σ_i',j'Being average and the variance of each pixel in the window of pixel centered by (i', j') respectively, δ is details decision threshold.

Compared with prior art, it has the great advantage that the present invention

Use and guide filter smoothing image, by adjusting its filtering core relevant parameter, it is thus possible to effectively distinguish image detail region and background area, while levels of detail is carried out self adaptation enhancing, effectively inhibit background noise, and amount of calculation is little, real-time, does not have gradient flop phenomenon.When Primary layer is processed, the inventive method, by the view data of rectangular histogram output is increased a bias factor, to successfully manage various scene, makes output picture contrast improve, and does not have the noise amplification that excessive tensile causes.

Accompanying drawing explanation

Fig. 1 is the inventive method flow chart.

Fig. 2 is the levels of detail view data row figure guiding wave filter to extract when taking different windows size and ε value, the horizontal from left to right each column of row figure represents when window size is 3 × 3,5 × 5 and 7 × 7 successively, the longitudinal each row from top to bottom of row figure represents when ε value is 100,500,1000 and 2000 successively, acquired levels of detail view data row figure.

Fig. 3 uses self adaptation to strengthen to strengthen, with non-self-adapting, the output image effect comparison diagram obtained respectively, wherein, Fig. 3 (a) is original image, Fig. 3 (b) uses non-self-adapting to strengthen the output image obtained, Fig. 3 (c) is that the self adaptation obtained in using this method process original image process strengthens gain coefficient mapping graph, and Fig. 3 (d) uses self adaptation in Fig. 3 (c) to strengthen the output image obtained after gain coefficient carries out details enhancing.

Fig. 4 uses the contained effect contrast figure obtained based on layered shaping frame algorithm of the inventive method and list of references two respectively, wherein, Fig. 4 (a) is original infrared image, Fig. 4 (b) is the process image using the inventive method to obtain, Fig. 4 (c) is the process image using two support methods of list of references to obtain, Fig. 4 (d) be in Fig. 4 (b) partial result 1. with the amplification effect comparison diagram 2. of partial result in Fig. 4 (c).

Detailed description of the invention

In conjunction with Fig. 1, the inventive method comprises the following steps:

Step one: use and guide wave filter that original infrared picture data is carried out the disposal of gentle filter, it is thus achieved that the base layer data of original infrared image, and base layer data and original infrared picture data are done difference, it is thus achieved that levels of detail data.

As a kind of preferred version of step one, described step one use shown in the computational methods such as formula (1) guiding wave filter to calculate base layer data,

$I_B(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}\left(a\right(i^{\′},j^{\′}\left)I_{in}\right(i^{\′},j^{\′})+b(i^{\′},j^{\′}\left)\right)---\left(1\right)$

In formula (1), I_inFor the original infrared picture data of input, I_BFor base layer data, (i, j) for pixel coordinate, w_i,jIt is that (w is window w for i, the window of pixel centered by j) with pixel_i,jThe number of middle pixel, a_(i',j')And b_(i',j')It is the linear coefficient centered by (i', j') in the window of pixel, a_(i',j')Shown in computational methods such as formula (2), b_(i',j')Shown in computational methods such as formula (3),

$a(i,j)=\frac{\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}{I}_{in}(i^{\′},j^{\′})G(i^{\′},j^{\′})-{\mathrm{\μ}}_{i,j}\stackrel{\‾}{I_{in}(i^{\′},j^{\′})}}{{\mathrm{\σ}}_{i,j}+\mathrm{\ϵ}}.---\left(2\right)$

$b(i,j)=\stackrel{\‾}{I_{in}(i^{\′},j^{\′})}-a(i,j){\mathrm{\μ}}_{i,j}.---\left(3\right)$

In formula (2) and formula (3), μ_i',j'Represent the average of each pixel in window；σ_i',j'Represent the variance of each pixel in window；ε is linear regression coeffficient, for determining the smoothness of this wave filter；Represent given window w_i,jMiddle I_inAverage；G (i', j') is the navigational figure data guiding wave filter for processing, G (i', j')=I_in(i, j), namely raw image data itself is as navigational figure data；

The value of size and ε by regulating the window guiding wave filter, can obtain detail extraction ability in various degree, and the inventive method guides the window size selection range of wave filter to be 3 × 3 to 9 × 9, and ε value is 100 to 2500.Accompanying drawing 2 gives under the value parameter combination of window size several groups different and ε, the levels of detail view data effect of extraction.

As the another kind of preferred version of step one, described base layer data and original infrared picture data are done shown in the computational methods such as formula (4) of difference acquisition levels of detail data,

I_D(i, j)=I_in(i,j)-I_B(i,j)(4)

In formula (4), I_DRepresent image detail layer data.

Step 2: use and guide filtering core that levels of detail data carry out self adaptation enhancing calculating, it is thus achieved that the enhanced levels of detail data of self adaptation.

As a kind of preferred version of step 2, described self adaptation strengthens shown in computational methods such as formula (5),

I_DP=I_D*(W(i,j)*a+b)(5)

In formula (5), I_DPThe levels of detail data obtained after calculating after strengthening for self adaptation, I_DFor image detail layer data, a and b is Serial regulation coefficient, and the span of a is 1.5 to 3, and the span of b is 0 to 0.5；W (i, is j) guide filter kernel coefficient, W (i, shown in the such as formula of computational methods j) (6),

$W(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\′},j^{\′})\∈w_{i,j}}{\Σ}\left(\frac{{\left(I\right(i,j)-{\mathrm{\μ}}_{i^{\′},j^{\′}})}^2}{{{\mathrm{\σ}}_{j^{\′},j^{\′}}}^2+\mathrm{\δ}}\right)---\left(6\right)$

In formula (6), (i is j) with pixel (i, the pixel value of the window of pixel each pixel interior, μ centered by j) to I_i',j'And σ_i',j'It is average and the variance of each pixel in the window of pixel centered by (i', j') respectively；δ is details decision threshold；(i, j) for as the mask differentiating details, levels of detail being carried out self adaptation enhancing for W.

Step 3: use histogram projection technology that base layer data carries out dynamic range compression and contrast enhancement processing, it is thus achieved that the base layer data after dynamic range compression.

A kind of preferred version as step 3, it is thus achieved that shown in the computational methods of the base layer data after dynamic range compression such as formula (7),

$I_{BP}=\frac{(D-R)}{P}+D\left(I_B\right)*R---\left(7\right)$

In formula (7), I_BPFor the base layer data after dynamic range compression；P is the regulatory factor for controlling output image overall brightness, and the brightness regulation factor can guarantee that when output gray level is less, and for ensureing that output image will not be compressed to very low brightness, the span of P can be further defined to 2 to 6；D is effective indication range of display, and D span is 0 to 255；D(I_B) for I_BAccumulation histogram data；R is the gray level maximum of output image；

D(I_B) shown in calculation such as formula (8), $D\left(I_B\right)=D\left(x\right)=\{\begin{array}{cc}0,& x=0\\ \frac{\underset{y=0}{\overset{x-1}{\Σ}}H\left(y\right)}{n_{vaalid}},& other\end{array}---\left(8\right)$

In formula (8), n_validRepresenting the valid gray level after projection process, H (y) is statistic histogram data, shown in H (y) value such as formula (9),

$H\left(y\right)=\{\begin{array}{cc}0,& n_y<T\\ 1,& n_y>T\end{array}---\left(9\right)$

In formula (9), n_yRepresent the gray level y number of pixels having；T is projection threshold value, and the span of T can be defined to 10～35 through a step；

Shown in the value mode of R such as formula (10),

R=min (n_valid,D)(10)。

Step 4: the base layer data after dynamic range compression described in enhanced for self adaptation described in step 2 levels of detail data and step 3 is merged, and rejects overflow value, it is thus achieved that finally export view data.

Described in step 4 shown in the final computational methods such as formula (11) exporting view data, (whether being necessary to be formulated again)

In formula (11), operatorRepresent and reject overflow value operation.

The effect of the inventive method can be further illustrated by following experimental result:

From accompanying drawing 3 it can be seen that use self adaptation to strengthen the output image obtained, background noise inhibition relatively use non-self-adapting strengthen the output image effect obtained good.

From accompanying drawing 4 it can be seen that directional arrow tiltedly in left side shows that Fig. 4 (b) suppresses background noise ratio effective；Right side tiltedly lower directional arrow shows that Fig. 4 (b) can retain original image information at strong adjacent edges, will not produce gradient upset effect.Therefore it being compared to two support methods of list of references, the inventive method can effectively strengthen original image overall contrast, strengthens image detail, can effectively suppress background noise, it does not have gradient flop phenomenon simultaneously.

Table 1 is the time contrast form using two support methods of list of references and the inventive method to process image, and test environment is Matlab2010b, and test view data is identical.As can be seen from Table 1, obtaining while good treatment effect, the inventive method amount of calculation is less, processes the time short, and treatment effeciency is high, and for comparing the occasions such as the hardware realization of valuing algorithm complex, this algorithm is with the obvious advantage.

This algorithm of table 1 uses time contrast (unit: second) with comparison algorithm

Claims (7)

1. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter, it is characterised in that comprise the following steps:

Step one: use and guide wave filter that original infrared picture data is carried out the disposal of gentle filter, it is thus achieved that the base layer data of original infrared image, and base layer data and original infrared picture data are done difference, it is thus achieved that levels of detail data；

Step 2: use and guide filtering core that levels of detail data carry out self adaptation enhancing calculating, it is thus achieved that the enhanced levels of detail data of self adaptation；

Step 3: use histogram projection technology that base layer data carries out dynamic range compression and contrast enhancement processing, it is thus achieved that the base layer data after dynamic range compression；

Step 4: the base layer data after dynamic range compression described in enhanced for self adaptation described in step 2 levels of detail data and step 3 is merged, and rejects overflow value, it is thus achieved that finally export view data；

Described step one use shown in the computational methods such as formula (1) guiding wave filter to calculate base layer data,

$I_B(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\&prime;},j^{\&prime;})\&Element;w_{i,j}}{\&Sigma;}\left(a\right(i^{\&prime;},j^{\&prime;}\left)I_{in}\right(i^{\&prime;},j^{\&prime;})+b(i^{\&prime;},j^{\&prime;}\left)\right)---\left(1\right)$

In formula (1), I_inFor the original infrared picture data of input, I_BFor base layer data, (i, j) for pixel coordinate, w_i,jIt is that (w is window w for i, the window of pixel centered by j) with pixel_i,jThe number of middle pixel, a_(i',j')And b_(i',j')It is the linear coefficient centered by (i', j') in the window of pixel, a_(i',j')Shown in computational methods such as formula (2), b_(i',j')Shown in computational methods such as formula (3),

$a(i,j)=\frac{\frac{1}{\left|w\right|}\underset{(i^{\&prime;},j^{\&prime;})\&Element;w_{i,j}}{\&Sigma;}{I}_{in}(i^{\&prime;},j^{\&prime;})G(i^{\&prime;},j^{\&prime;})-{\mathrm{\&mu;}}_{i,j}\stackrel{\&OverBar;}{I_{in}(i^{\&prime;},j^{\&prime;})}}{{\mathrm{\&sigma;}}_{i,j}+\mathrm{\&epsiv;}}.---\left(2\right)$

$b(i,j)=\stackrel{\&OverBar;}{I_{in}(i^{\&prime;},j^{\&prime;})}-a(i,j){\mathrm{\&mu;}}_{i,j}.---\left(3\right)$

In formula (2) and formula (3), μ_i',j'Represent the average of each pixel in window；σ_i',j'Represent the variance of each pixel in window；ε is linear regression coeffficient, for determining the smoothness of this wave filter；Represent given window w_i,jMiddle I_inAverage；G (i', j') is the navigational figure data guiding wave filter for processing, G (i', j')=I_in(i, j), namely raw image data itself is as navigational figure data；

In described step one, base layer data and original infrared picture data are done shown in the computational methods such as formula (4) of difference acquisition levels of detail data,

I_D(i, j)=I_in(i,j)-I_B(i,j)(4)

In formula (4), I_DRepresent image detail layer data；

In described step 2, self adaptation strengthens shown in computational methods such as formula (5),

I_DP=I_D*(W(i,j)*a+b)(5)

In formula (5), I_DPThe levels of detail data obtained after calculating after strengthening for self adaptation, a and b is Serial regulation coefficient, W (i, is j) guide filter kernel coefficient, W (i, shown in the such as formula of computational methods j) (6),

$W(i,j)=\frac{1}{\left|w\right|}\underset{(i^{\&prime;},j^{\&prime;})\&Element;w_{i,j}}{\&Sigma;}\left(\frac{{\left(I\right(i,j)-{\mathrm{\&mu;}}_{i^{\&prime;},j^{\&prime;}})}^2}{{{\mathrm{\&sigma;}}_{i^{\&prime;},j^{\&prime;}}}^2+\mathrm{\&delta;}}\right)---\left(6\right)$

In formula (6), (i is j) with pixel (i, the pixel value of the window of pixel each pixel interior, μ centered by j) to I_i',j'And σ_i',j'Being average and the variance of each pixel in the window of pixel centered by (i', j') respectively, δ is details decision threshold.

2. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 1, it is characterised in that the window size selection range of described guiding wave filter is 3 × 3 to 9 × 9, and ε span is 100 to 2500.

3. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 1, it is characterised in that the span of a is 1.5 to 3, and the span of b is 0 to 0.5.

4. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 1, it is characterised in that described step 3 obtains shown in the computational methods such as formula (7) of the base layer data after dynamic range compression,

$I_{BP}=\frac{(D-R)}{P}+D\left(I_B\right)*R---\left(7\right)$

In formula (7), I_BPFor the base layer data after dynamic range compression；P is the regulatory factor for controlling output image overall brightness；D is effective indication range of display, and D span is 0～255；D(I_B) for I_BAccumulation histogram data；R is the gray level maximum of output image；

D(I_B) shown in calculation such as formula (8), $D\left(I_B\right)=D\left(x\right)=\left\{\begin{array}{cc}0,& x=0\\ \frac{\underset{y=0}{\overset{x-1}{\&Sigma;}}H\left(y\right)}{n_{valid}},& other\end{array}\right.---\left(8\right)$

In formula (8), n_validRepresenting the valid gray level after projection process, H (y) is statistic histogram data, shown in H (y) value such as formula (9),

$H\left(y\right)=\left\{\begin{array}{cc}0,& n_y<T\\ 1,& n_y>T\end{array}\right.---\left(9\right)$

In formula (9), n_yRepresenting the gray level y number of pixels having, T is projection threshold value；

Shown in the value mode of R such as formula (10),

R=min (n_valid,D)(10)。

5. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 4, it is characterised in that the span of described brightness regulation factor P is 2 to 6.

6. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 4, it is characterised in that the span 10 to 35 of described projection threshold value T.

7. show and detail enhancing method based on the Larger Dynamic scope infrared image guiding wave filter as claimed in claim 1, it is characterised in that described in step 4 shown in the final computational methods such as formula (11) exporting view data,

In formula (11), operatorRepresent and reject overflow value operation.