CN105205794B - A kind of synchronous enhancing denoising method of low-light (level) image - Google Patents

Abstract

The invention discloses a method for synchronous enhancement and denoising of low-illuminance images, which uses a foggy image degradation model considering noise and combines iterative joint bilateral filtering to simultaneously perform contrast enhancement and noise removal operations on low-illuminance images; in the iterative In the initialization stage, the dark channel prior theory is used to obtain the initial estimated values of the global atmospheric light, transmittance and scene light of the inverted image of the low-illumination image; The result of the iterative joint bilateral filtering method is used for detail compensation through the quotient image; in the first round of iteration, the guided image is the noise image itself, and in the subsequent iteration process, the result of each round of iteration is used as the next iteration guide map; finally, the enhanced restoration image is obtained by inverting the iterative result again; the invention can simultaneously improve the low-light image visualization effect and remove image noise, and has a good visual effect.

Description

A Synchronous Enhancement Denoising Method for Low Illumination Images

technical field

The invention relates to the technical field of image enhancement and denoising, in particular to a synchronous enhancement and denoising method for low-illuminance images

Background technique

In a low-light environment, the images obtained by the image acquisition equipment not only have low recognizability, but also contain a lot of noise. The image degradation caused by the low-light environment not only affects the recognition of the image by the human eye, but also makes intelligent transportation, video surveillance and target recognition difficult. The performance of computer vision systems such as computer vision is greatly affected, so it is of great value to enhance and denoise low-light images; because the gray coverage of images collected in low-light environments is very narrow, and the pixel values are at a low level , so the main purpose of enhancing low-light images is to expand the gray range of the image and increase the overall brightness of the image, so that the image information that cannot be discerned can be recognized by human eyes or machines; traditional image enhancement methods can be divided into spatial There are two types of domain enhancement methods and frequency domain enhancement methods: histogram equalization is a classic spatial domain enhancement method, which can effectively enhance image contrast, but this method may cause the original brighter pixels to oversaturate and lose image structure information ; frequency domain enhancement methods such as wavelet transform, by converting the image signal to the frequency domain, the wavelet coefficients are processed to achieve the enhanced effect; the Retinex method is an image enhancement method based on the retinal cerebral cortex theory, which divides the image into There are two parts of the brightness map and the reflection map. The enhanced effect is achieved by reducing the influence of the brightness map on the reflection map. With the introduction of the Retinex theory, a large number of researchers have successively proposed related improved algorithms, such as single-scale Retinex and multi-scale Retinex. , multi-scale Retinex with color factor restoration, etc., these algorithms can achieve certain enhancement effects.

In recent years, the low-illumination image enhancement method based on defogging technology has opened up a new way of low-illumination enhancement. By comparing the similarity between the inverted image of the low-illumination image and the foggy image, the inverted low-illuminance video frame is used The dark primary color defogging method can get better visual effects. Low-illumination images not only have low gray levels but also have high noise content, and most enhancement methods will amplify the noise while performing intensity conversion; At present, there are many researches that can complete image enhancement and denoising separately, but there are few enhancement and denoising methods for low-illuminance image features. However, because the dark primary color dehazing technology does not consider the influence of noise, the noise is significantly amplified after enhancement.

The Chinese invention patent with the application number CN201510260607 discloses a video monitoring and acquisition system with image enhancement function, including an image acquisition device for acquiring images, an analog-to-digital conversion device AD, and a digital image signal converted by analog-to-digital conversion. An image processing module for enhanced processing, a digital-to-analog conversion device DA, a video storage module, a monitor terminal, and a control center module for controlling the operation status of the entire system. The video monitoring and acquisition system of the invention adopts an embedded hardware platform, and The software algorithm is optimized, so that the size and power consumption of the image processing hardware equipment are greatly reduced, real-time image enhancement can be realized, and the integration with the image acquisition equipment is convenient. The modular design is also conducive to the free collocation of functional modules, which is more It is suitable for the application and reduces the purchase cost.

The Chinese invention patent with the application number CN201510329457 discloses a grayscale image enhancement method based on the retinal mechanism. The specific process includes estimating the adaptive parameters of the global brightness determination algorithm, generating the brightness map of the image, calculating the brightness enhanced image and edge enhancement processing, Firstly, the adaptive parameters are estimated through the brightness distribution of the global dark area; then the global brightness enhancement processing is performed on the image respectively, and the modulation map of the whole picture is obtained by the modulation function, and the brightness enhancement result is calculated; Finally, the edge enhancement is realized based on the adaptive scale Gaussian difference model. The model scale is affected by the contrast, and finally the finer texture information can be enhanced in the bright area, and the larger outline information can be enhanced in the dark area.

Although the above disclosed inventions can effectively enhance the contrast of images, they do not take into account the influence of noise conditions on image enhancement. Therefore, creative improvements to the prior art are required.

Contents of the invention

The purpose of the present invention is to provide a synchronously enhanced denoising method for low-illuminance images, which utilizes the noise-considered fog map degradation model and iterative joint bilateral filtering algorithm to simultaneously perform contrast-enhancing operations and noise removal operations for low-illuminance images, thereby effectively Improve image contrast and suppress noise to enhance the visual effect of the image.

To achieve the above object, the present invention provides the following technical solutions:

A synchronous enhancement denoising method for low-illumination images. The low-illuminance image enhancement algorithm is an enhancement algorithm based on dark channel color dehazing technology. The synchronous enhancement denoising operation is completed by iterative joint bilateral filtering and alternately correcting the fog image degradation model parameters, including the following step:

1) Input the original low-light image I( _xi ) into the computer image processing system, and invert it to obtain the inverted image I _inv ( _xi );

2) Calculate the global atmospheric light value A ^c in the inverted image I _inv ( _xi ) according to the dark channel prior theory;

3) Calculate the initial transmittance t ⁰ of the image according to the brightness map of the inverted image I _inv ( _xi );

4) Substitute the global atmospheric light A ^c obtained in step 2 and the initial transmittance t ⁰ obtained in step 3 into the foggy image degradation model to obtain the initial estimated value of scene light

5) Use the iterative joint bilateral filtering method to alternately modify the parameters in the foggy image degradation model, and use the quotient image method to compensate the details for the results of each round.

6) Use the scene light finally obtained in step 5 Inversion is performed to obtain the final enhanced denoising result.

As a further solution of the present invention: in the step 1, for the input low-illuminance image When performing a reversal operation, the reversal algorithm is as follows:

Among them, I represents the original input low-light image, I _inv represents the inverted image, and c represents a color channel in the RGB three color channels of the image.

As a further solution of the present invention: said step 2 includes the following steps:

a) Perform minimum value filtering on each color channel of the inverted image I _inv ( _xi ), and calculate the minimum value of the three-channel filtering results for each pixel as the dark primary color value of the pixel, thereby obtaining an inverted image The dark primary color map;

b) Select 0.1% of the pixels with the largest intensity value among all the pixels in the dark primary color map, and mark the positions of these pixels, corresponding to the positions in the three color channels of the inverted image I _inv ( _xi ), Find the intensity value of the brightest point of each channel as the atmospheric light A ^c of the color channel.

As a further solution of the present invention ^: in the step 3, the algorithm for obtaining the initial transmittance t is as follows:

t ⁰ ( _xi )=CY( _xi )

In the formula, C is a parameter used to weaken the brightness image Y, and the value range of C is [1.06, 1.08], and the calculation method of the brightness image Y( _xi ) is as follows:

Y( _xi )＝0.299×R+0.587×G+0.114×B

In the formula: R, G, and B respectively represent the RGB three-channel component values of the image.

As a further solution of the present invention: in the step 4, the algorithm for obtaining the initial estimated value of the scene light is as follows

where I _inv ( _xi ) is the inverse image of the input image, t ⁰ ( _xi ) is the initial estimated value of transmittance, A ^c is the global atmospheric light, is the lower limit of the transmittance, usually 0.01.

As a further solution of the present invention: the form of the foggy image degradation model considering noise used in the step 5 is as follows:

I _inv (x)＝J _inv (x)t(x)+A(1-t(x))+n(x)

As a further solution of the present invention: in the iterative joint bilateral filtering method in the step 5, during the first round of iteration, the guide map is the noise image itself, and in the subsequent iteration process, the result of each round of iteration is used as the next iteration A directed graph for round iterations.

As a further solution of the present invention: the step 5 includes the following steps:

a) Set the initial values of transmittance and scene light in the iterative process as t ⁰ and The number of collocation iterations k=1;

b) Use the corrected scene light from the previous iteration Correct the transmittance t ^k in this round of iterations, and the algorithm for correcting the transmittance is as follows:

where g _d ( _xi -x) is the spatial domain kernel function, is the range kernel function, Ω(x) is the neighborhood centered on x;

c) Use the corrected transmittance value t ^k in the current iteration to correct the scene light in the current iteration The algorithm for correcting scene light is as follows:

In the formula, g _d ( _xi -x) space domain kernel function, is the range kernel function, Ω(x) is the neighborhood centered on x;

d) Use the quotient image to compensate the details of each round of filtering results, and obtain the final scene light correction result of this round of iterations The detail compensation algorithm is as follows:

In the formula, M is used to balance the weight of the detail image, is the quotient image, the calculation method is as follows:

As a further solution of the present invention: in step 5, the filter window is selected to be 15×15 when the transmittance is corrected, and the filter window is selected to be 7×7 when the scene light is corrected.

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

(1) The present invention proposes a synchronously enhanced denoising method for low-illuminance images, using a fog map degradation model considering noise and an iterative joint bilateral filtering algorithm to simultaneously perform contrast enhancement and noise removal operations for low-illuminance images, thereby Effectively improve the image contrast and suppress noise, enhance the visual effect of the image;

(2) In the iterative process of each round of enhanced denoising, the present invention uses the quotient image to compensate the iterative result, so that the processing result contains rich detail information;

(3) The present invention fully considers the characteristics of the low-illumination image, establishes a physical model for the characteristics of the low-illumination image, and simultaneously completes the enhancement and denoising operations.

Description of drawings

Fig. 1 is the flow chart of low-illuminance image synchronous enhancement denoising algorithm of the present invention;

Figure 2 is the unprocessed photo.

Fig. 3 is a processing effect diagram of Fig. 2 using the low-illuminance image synchronously enhanced denoising algorithm of the present invention.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Please refer to Figure 1, a synchronous enhancement denoising method for low-illumination images, first using the dehazing technique based on the dark channel prior theory to estimate the initial estimates of the global atmospheric light, transmittance and scene light of the inverted image of the low-illumination image ; Then through iterative joint bilateral filtering to alternately modify the unknown parameters in the fog map degradation model to achieve synchronous enhanced denoising, and finally reverse the image to obtain the final enhanced result, including the following steps:

1) Input the original low-illuminance image I( _xi ) into the computer image processing system, and invert it to obtain the inverted image I _inv ( _xi ), for the input low-illuminance image When performing a reversal operation, the reversal algorithm is as follows:

2) According to the dark channel prior theory, calculate the global atmospheric light value A ^c in the inverted image I _inv ( _xi ), the specific steps are as follows:

a) Perform minimum value filtering on each color channel of the inverted image I _inv ( _xi ), and calculate the minimum value of the three-channel filtering results for each pixel as the dark primary color value of the pixel, thereby obtaining an inverted image The dark primary color map;

b) Select 0.1% of the pixels with the largest intensity value among all the pixels in the dark primary color map, and mark the positions of these pixels, corresponding to the positions in the three color channels of the inverted image I _inv ( _xi ), Find the intensity value of the brightest point of each channel as the atmospheric light A ^c of the color channel;

3) Calculate the initial transmittance t ⁰ of the image according to the brightness map of the inverted image I _inv ( _xi ), and the algorithm for calculating the initial transmittance t ⁰ is as follows:

t ⁰ ( _xi )=CY( _xi )

In the formula, C is a parameter used to weaken the brightness image Y, and the value range of C is [1.06, 1.08], and the calculation method of the brightness image Y( _xi ) is as follows:

Y( _xi )＝0.299×R+0.587×G+0.114×B

In the formula: R, G, and B respectively represent the RGB three-channel component values of the image.

4) Substitute the global atmospheric light A ^c obtained in step 2 and the initial transmittance t ⁰ obtained in step 3 into the foggy image degradation model to obtain the initial estimated value of scene light The algorithm for obtaining the initial estimated value of the scene light is as follows:

where I _inv ( _xi ) is the inverse image of the input image, t ⁰ ( _xi ) is the initial estimated value of transmittance, A ^c is the global atmospheric light, is the lower limit of transmittance, usually 0.01;

5) Use the iterative joint bilateral filtering method to alternately modify the parameters in the foggy image degradation model considering noise, and use the quotient image method to perform detail compensation on the results of each round. The specific steps are as follows:

a) Set the initial values of transmittance and scene light in the iterative process as t ⁰ and The number of collocation iterations k=1;

b) Use the corrected scene light from the previous iteration Correct the transmittance t ^k in this round of iterations, and the algorithm for correcting the transmittance is as follows:

where g _d ( _xi -x) is the spatial domain kernel function, is the range kernel function, Ω(x) is the neighborhood centered on x;

c) Use the corrected transmittance value t ^k in the current iteration to correct the scene light in the current iteration The algorithm for correcting scene light is as follows:

In the formula, g _d ( _xi -x) space domain kernel function, is the range kernel function, Ω(x) is the neighborhood centered on x;

d) Use the quotient image to compensate the details of each round of filtering results, and obtain the final scene light correction result of this round of iterations The detail compensation algorithm is as follows:

In the formula, M is used to balance the weight of the detail image, is the quotient image, the calculation method is as follows:

In the iterative joint bilateral filtering method in the step 5, during the first round of iteration, the guide map is the noise image itself, and in the subsequent iteration process, the result of each round of iteration is used as the guide map of the next round of iteration, And when correcting the transmittance, the filter window is selected to be 15×15, and when the scene light is corrected, the filter window is selected to be 7×7;

6) Use the scene light finally obtained in step 5 Inversion is performed to obtain the final enhanced denoising result.

Example:

The present invention provides a synchronously enhanced denoising method for low-illuminance images, which effectively removes image noise while improving the visualization effect of low-illuminance images. The effect of the present invention can be further illustrated by the following experimental data:

Please refer to Figure 2-3, Figure 2 is a low-illuminance image with a size of 1920×1080 and Gaussian white noise with a standard deviation of 5, the final result of processing using the synchronization enhancement denoising method proposed by the present invention is shown in Figure 3 As shown in Figure 2-3, it can be seen that the low-illuminance synchronous enhancement and denoising method proposed by the present invention can effectively improve the overall brightness of the image and effectively remove image noise. In the enhancement results, not only can you observe originally invisible scenes such as steps information, and details such as license plate numbers are still clearly visible and unaffected by the denoising operation.

The present invention proposes a synchronously enhanced denoising method for low-illuminance images, which uses the noise-considered fog map degradation model and iterative joint bilateral filtering algorithm to simultaneously perform contrast enhancement and noise removal operations on low-illuminance images, thereby effectively improving the image Contrast and suppress noise, enhance the visual effect of the image; in each round of iterative process of enhanced denoising, use the quotient image to compensate the iterative results, so that the processing results contain rich detailed information; fully consider the characteristics of low-light images, A physical model is established for the characteristics of low-illumination images, and the enhancement and denoising operations are completed simultaneously.

It will be apparent to those skilled in the art that the invention is not limited to the details of the above-described exemplary embodiments, but that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Accordingly, the embodiments should be regarded in all points of view as exemplary and not restrictive, the scope of the invention being defined by the appended claims rather than the foregoing description, and it is therefore intended that the scope of the invention be defined by the appended claims rather than by the foregoing description. All changes within the meaning and range of equivalents of the elements are embraced in the present invention. Any reference sign in a claim should not be construed as limiting the claim concerned.

In addition, it should be understood that although this specification is described according to implementation modes, not each implementation mode only includes an independent technical solution, and this description in the specification is only for clarity, and those skilled in the art should take the specification as a whole , the technical solutions in the various embodiments can also be properly combined to form other implementations that can be understood by those skilled in the art.

Claims (7)

1. the synchronous enhancing denoising method of a kind of low-light (level) image, it is characterised in that comprise the following steps：

1) by original low-light (level) image I (x_i) input Computerized image processing system, and inverted, the image after being inverted I_inv(x_i)；

2) according to dark primary priori theoretical, reverse image I is asked for_inv(x_i) in global air light value A^c；

3) according to reverse image I_inv(x_i) luminance graph calculate image initial transmission t⁰；

4) the global atmosphere light A for trying to achieve step 2^cThe initial transmission t tried to achieve with step 3⁰Misty Image degradation model is substituted into obtain To the initial estimate of scene light

5) parameter in the Misty Image degradation model for considering noise is alternately corrected using the joint bilateral filtering method of iteration, and Result to each round uses the progress details compensation of quotient image method；The greasy weather figure of the consideration noise used in the step 5 As the form of degradation model is as follows：

I_inv(x)=J_inv(x)t(x)+A(1-t(x))+n(x)

The joint bilateral filtering method of iteration in the step 5 is in first round iteration, and it is noise image itself to be oriented to figure, In later iterative process, using the result of each round iteration as next round iteration guiding figure；

The step 5 comprises the following steps implementation：

A) initial value for setting transmissivity and scene light in iterative process is respectively t⁰WithJuxtaposition iterations k=1；

B) using revised scene light in last round of iterationCorrect the transmissivity t in epicycle iteration^k, amendment transmissivity Algorithm is as follows：

<mrow> <msup> <mi>t</mi> <mi>k</mi> </msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>&amp;Omega;</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </munder> <msub> <mi>g</mi> <mi>d</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>-</mo> <mi>x</mi> <mo>)</mo> </mrow> <msub> <mi>g</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>t</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <msup> <mi>t</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> </msub> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <mi>A</mi> <mo>)</mo> </mrow> </mrow> <mrow> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>&amp;Omega;</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </munder> <msub> <mi>g</mi> <mi>d</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>-</mo> <mi>x</mi> <mo>)</mo> </mrow> <msub> <mi>g</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>t</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <msup> <mi>t</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <mi>A</mi> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>

In formula, g_d(x_i- x) it is spatial domain kernel function, g_r(t^k-1(x_i)-t^k-1(x) it is) codomain kernel function, Ω (x) is centered on x Neighborhood；

C) using revised transmittance values t in epicycle iteration^kCorrect the scene light in epicycle iterationCorrect the calculation of scene light Method is as follows：

<mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mi>k</mi> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>&amp;Omega;</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </munder> <msub> <mi>g</mi> <mi>d</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>-</mo> <mi>x</mi> <mo>)</mo> </mrow> <msub> <mi>g</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> </msub> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <mi>A</mi> <mo>(</mo> <mrow> <mn>1</mn> <mo>-</mo> <msup> <mi>t</mi> <mi>k</mi> </msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mo>)</mo> <mo>)</mo> </mrow> </mrow> <mrow> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>&amp;Omega;</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </munder> <msub> <mi>g</mi> <mi>d</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>-</mo> <mi>x</mi> <mo>)</mo> </mrow> <msub> <mi>g</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>-</mo> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>)</mo> </mrow> <msup> <mi>t</mi> <mi>k</mi> </msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>

In formula, g_d(x_i- x) spatial domain kernel function,For codomain kernel function, Ω (x) is centered on x Neighborhood；

D) detail section of each round filter result is compensated using quotient image, obtains the final scene light of epicycle iteration Correction resultDetails backoff algorithm is as follows：

<mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mi>F</mi> <mi>i</mi> <mi>n</mi> <mi>a</mi> <mi>l</mi> </mrow> </msubsup> <mo>=</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>M</mi> <mo>)</mo> </mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mi>D</mi> <mi>e</mi> <mi>t</mi> <mi>a</mi> <mi>i</mi> <mi>l</mi> </mrow> </msubsup> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mi>k</mi> </msubsup> <mo>+</mo> <msubsup> <mi>MJ</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mi>k</mi> </msubsup> </mrow>

In formula, M is used to balance the weight shared by detail pictures,For quotient image, computational methods are as follows：

<mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mi>D</mi> <mi>e</mi> <mi>t</mi> <mi>a</mi> <mi>i</mi> <mi>l</mi> </mrow> </msubsup> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>+</mo> <mi>e</mi> </mrow> <mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mi>k</mi> </msubsup> <mo>+</mo> <mi>e</mi> </mrow> </mfrac> </mrow>

In formula, ε is used for the influence for weakening noise；

6) by the scene light finally obtained in step 5Inverted, obtain final enhancing denoising result.

2. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that the scene light Initial estimateAsk for carried out simultaneously with noise remove.

3. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that in the step 1, To input low-light (level) imageWhen carrying out reverse turn operation, specific reversion algorithm is as follows：

<mrow> <msubsup> <mi>I</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mi>c</mi> </msubsup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mn>255</mn> <mo>-</mo> <msup> <mi>I</mi> <mi>c</mi> </msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow>

Wherein, I represents the original low-light (level) image of input, I_invRepresent in reverse image, the Color Channels of c representative images RGB tri- One Color Channel.

4. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that the step 2 is wrapped Include following steps implementation：

A) to reverse image I_inv(x_i) each Color Channel do mini-value filtering, and each pixel is asked for triple channel filtering As a result minimum value as the pixel dark primary value, so as to obtain the dark primary figure of reverse image；

B) 0.1% maximum pixel of intensity level in all pixels point in dark primary figure is chosen, by the position of these pixels It is marked, in reverse image I_inv(x_i) position corresponding in three Color Channels, find the strong of each passage most bright point Angle value as the Color Channel atmosphere light A^c。

5. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that in the step 3, Ask for initial transmission t⁰Algorithm it is as follows：

t⁰(x_i)=C-Y (x_i)

In formula, C is the parameter for weakening luminance picture Y, and C spans are [1.06,1.08], luminance picture Y (x_i) it is specific Calculation is as follows：

Y(x_i)=0.299 × R+0.587 × G+0.114 × B

In formula：R, G, B distinguish representative image RGB triple channel component values.

6. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that in the step 4, The algorithm for asking for scene light initial estimate is as follows：

<mrow> <msubsup> <mi>J</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> <mn>0</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>I</mi> <mrow> <mi>i</mi> <mi>n</mi> <mi>v</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msup> <mi>A</mi> <mi>c</mi> </msup> </mrow> <mrow> <mi>max</mi> <mrow> <mo>(</mo> <msup> <mi>t</mi> <mn>0</mn> </msup> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> <mo>,</mo> <mover> <mi>t</mi> <mo>^</mo> </mover> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>+</mo> <msup> <mi>A</mi> <mi>c</mi> </msup> </mrow>

In formula, I_inv(x_i) be input picture reverse image, t⁰(x_i) be transmissivity initial estimate, A^cFor global atmosphere light,For the lower limit of transmissivity, 0.01 is generally taken.

7. the synchronous enhancing denoising method of low-light (level) image according to claim 1, it is characterised in that in the step 5 Filter window size chooses 15 × 15 when correcting transmissivity, and filter window size chooses 7 × 7 during amendment scene light.