CN104732227A - Rapid license-plate positioning method based on definition and luminance evaluation - Google Patents

Abstract

The invention discloses a rapid license-plate positioning method based on definition and luminance evaluation. The rapid license-plate positioning method includes the following steps: 1, carrying out definition evaluation based on noise and sharpness on an input image; 2, carrying out gradient sharpening processing if the definition is insufficient, and carrying out Gaussian fuzzy processing if the definition is excessively high; 3, converting the RGB image into a grey-scale image; 4, carrying out luminance evaluation on the grey-scale image; 5, carrying out illumination unification processing if the luminance is abnormal; 6, extracting the perpendicular edges in the image through a Scharr operator; 7, carrying out local self-adaption threshold processing; 8, filtering the noise through morphological processing, and fusing the perpendicular edges to form a communicated area; 9, carrying out area marking on the communicated area, carrying out screening according to the characteristics of Chinese license plates, and obtaining a license plate area.

Description

A kind of Location Method of Vehicle License Plate based on sharpness and luminance evaluation

Technical field

The invention belongs to image processing field, particularly a kind of Location Method of Vehicle License Plate based on sharpness and luminance evaluation.

Background technology

In intelligent transportation system, the number-plate number recognition technology based on digital picture has become indispensable important component part and has been widely used in the fields such as monitoring, alarming, punishment on contravention of regulation, room entry/exit management and expressway tol lcollection management.License Plate, namely orients the position of car plate in image or video, and be the first step of number-plate number identification, its accuracy directly affects the effect of number-plate number identification, is also a step the most consuming time simultaneously.Therefore License Plate has critical role in intelligent transportation system.

Current license plate locating method is divided three classes:

1, based on the method for marginal information.The method is most widely used general, utilizes the more feature of car plate vertical edge to position.For simple scenario as room entry/exit management and charge station etc., there is good verification and measurement ratio and real-time, but for the scene of complexity as highway etc., because car plate is comparatively fuzzy and illumination is uncontrollable, cause the accuracy rate of License Plate low.

2, based on the method for colouring information.The method utilizes known car plate colouring information to assist the marginal information of car plate to position, can significantly raising accuracy rate for the single scene of light conditions, but retroaction can be caused such as night etc. in illumination most complex scenarios, and also higher for image resolution requirement, be only applicable to the Vehicle License Plate Recognition System in the few scene of reality.

3, based on the method for machine learning.The method by the car plate sample under the different scene of collection and for training the world model comprising different situation in sample, and then is generalized to the unknown situation in new images.For the sharpness of new images or brightness ideal close to effect during training sample, but when difference is larger, then accuracy rate is low and be difficult to ensure real-time.

In sum, based on the license plate locating method of marginal information because the advantages such as operand is little, processing speed is fast, required storage space is little are by generally application is with Real-time Vehicle License Plate automatic recognition system, but its shortcoming major embodiment is following 2 points: 1, can not make adjustment according to the concrete condition of input data; 2, under complex scene, accuracy rate is low, poor real.

Summary of the invention

Fundamental purpose of the present invention is that the shortcoming overcoming prior art is with not enough, a kind of Location Method of Vehicle License Plate based on sharpness and luminance evaluation is provided, the method efficiently solves car plate and is difficult to be accurately positioned under different sharpness and complex illumination environment or can not orientation problem, thus improves the accuracy rate of License Plate.

In order to arrive above-mentioned purpose, the present invention by the following technical solutions:

Based on a Location Method of Vehicle License Plate for sharpness and luminance evaluation, comprise the steps:

(1) intelligibility evaluation based on noise and acutance is carried out to input picture;

(2) carry out Grads Sharp process to the image of sharpness deficiency, the too high image of sharpness carries out Gaussian Blur process;

(3) by RGB figure, gray-scale map is converted to the image after step (2) process;

(4) luminance evaluation is carried out to gray-scale map, namely too high the or luminance shortage two kinds of situations of brightness are carried out to gray level image and assess;

(5) if brightness is abnormal, illumination normalization is carried out;

(6) by vertical edge in Scharr operator extraction image;

(7) local auto-adaptive threshold process is carried out;

(8) by Morphological scale-space filtered noise, and vertical edge is made to be fused into connected region;

(9) zone marker is carried out to connected region, carry out screening and obtaining license plate area according to the feature of Chinese car plate.

Preferably, in step (1), intelligibility evaluation is made up of noise evaluation and assess sharpness, after the noise evaluation value obtaining image and assess sharpness value, obtained the intelligibility evaluation value of image by certain weighted connections, the formula calculating intelligibility evaluation value is as follows:

Quality＝wNoise+(1-w)Sharpness

Wherein, w is weighted value, and span between zero and one;

Described noise evaluation obtains binary map by first carrying out OTSU threshold process to image, then calculates area in connected region and be less than the accounting of given threshold value, obtain noise evaluation value finally by linear transformation.The formula of calculating noise assessed value is as follows:

$\mathrm{Noise}=a\underset{{\mathrm{Area}}_i=0}{\overset{T_{\mathrm{area}}}{\mathrm{\Σ}}}{\mathrm{Comp}}_i/\underset{{\mathrm{Area}}_i=0}{\overset{\∞}{\mathrm{\Σ}}}{\mathrm{Comp}}_i+b$

Wherein, Comp _ii-th region, Area _ithe area in i-th region, T _areabe the given threshold value judging that whether area is too small, a and b is the parameter of linear transformation;

What described assess sharpness adopted is point sharpness method, gets 8 neighborhood points and subtracts each other with it, first ask the weighted sum of 8 differences to often of image, by have an income value to be added divided by the total number of pixel, the formula calculating assess sharpness value is as follows:

$\mathrm{Sharpness}=\underset{i=1}{\overset{m\×n}{\mathrm{\Σ}}}\underset{a=1}{\overset{8}{\mathrm{\Σ}}}\left|\frac{\mathrm{df}}{\mathrm{dx}}\right|/(m\×n)$

Wherein, a is a field pixel, m and n is the length of image and wide, and df is the change amplitude of gray scale, and dx is the distance increment between pixel.

Preferably, step (2) is specially:

(2-1) given two threshold value T _quality1and T _quality2, wherein T _quality1the threshold value judging that whether image definition is not enough, T _quality2it is the threshold value judging that whether image definition is too high;

(2-2) as Quality < T _quality1time Grads Sharp process is carried out to image, namely ask for image gradient information Grad by Laplace operator, be then added with former figure and obtain Grads Sharp result, the formula of Grads Sharp is as follows:

f(x，y)＝f(x，y)+Grad(x，y)

Wherein, f (x, y) for former figure, Grad (x, y) be gradient information;

(2-3) as Quality > T _quality2time Gaussian Blur process is carried out to image, namely by using Gaussian function to calculate in the Fuzzy Template of normal distribution, and do convolution algorithm with former figure and obtain Gaussian Blur result, the formula of Gaussian function and Gaussian Blur is as follows:

$\mathrm{Gauss}(a,b)=\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}{e}^{-\frac{{(a-m/2)}^2+{(b-n/2)}^2}{2{\mathrm{\&sigma;}}^2}}$

f(x，y)＝f(x，y)*Gauss(a，b)

Wherein, σ is the standard deviation of normal distribution, m and n is that template is long and wide, a and b is element in template, and f (x, y) is former figure.

Preferably, in step (3), the formula being turned gray-scale map by RGB figure is:

Gray(x，y)＝0.299R(x，y)+0.587G(x，y)+0.114B(x，y)；

Or Gray (x, y)=(38R (x, y)+75G (x, y)+15B (x, y)) > > 7;

Or $\mathrm{Gray}(x,y)={(0.2973R{(x,y)}^{2.2}+0.6274G{(x,y)}^{2.2}+0.0753B{(x,y)}^{2.2})}^{\frac{1}{2.2}}$ Wherein, R, G, B are the value of red, green, blue three Color Channels of pixel (x, y), and Gray is the gray-scale value of pixel (x, y) correspondence.

Preferably, in step (4), the method for gray scale being carried out to luminance evaluation is:

Calculate pixel value lower than threshold value T _bright1or higher than threshold value T _bright2pixel account for the ratio sum of all pixels, if be greater than given threshold value T _bright3then thinking, is that brightness is abnormal.

Preferably, in step (5), the method for illumination normalization is:

(5-1) first correct enhancing brightness of image with Gamma, Gamma updating formula is as follows:

g(x，y)＝c[f(x，y)]γ

In formula, f (x, y) is input picture, and g (x, y) is output image, c be arbitrary value for adjusting picture contrast, γ between 0 ~ 1 for adjusting brightness of image;

(5-2) then carry out difference of Gaussian filtering and remove illumination high frequency components, i.e. uneven part, difference of Gaussian Filtering Formula is as follows:

g(x，y)＝f(x，y)*Gauss(σ ₂)-f(x，y)*Gauss(σ ₁)

In formula, f (x, y) is input picture, and g (x, y) is output image, Gauss (σ _n) for variance be σ _nthe Gaussian function of (n=1,2);

(5-3) finally adopt contrast equalization to strengthen the contrast of image irradiation, thus at utmost retain the useful details of image and strengthen picture contrast while of reaching uneven in removal illumination, contrast equalization formula is as follows:

$g_1(x,y)=\frac{f(x,y)}{{\left(\mathrm{mean}\left({\left|f(x^{\&prime;},y^{\&prime;})\right|}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_2(x,y)=\frac{g_1(x,y)}{{\left(\mathrm{mean}\left(\min {(\mathrm{\&tau;},|g_1(x^{\&prime;},y^{\&prime;})\left|\right)}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_3(x,y)=\tanh \left(\frac{g_2(x,y)}{\mathrm{\&tau;}}\right)$

In formula, f (x, y) is input picture, g _n(x, y) (n=1,2,3) are output images, and α is the impact of compressibility index for reducing large pixel value, and τ amputates large pixel value in image after being used for first step normalization.

Preferably, in step (6), by the concrete grammar of vertical edge in Scharr operator extraction image be:

Scharr template is expressed as follows:

${\mathrm{Grad}}_x=\left[\begin{array}{ccc}-3& 0& +3\\ -10& 0& +10\\ -3& 0& +3\end{array}\right]{\mathrm{Grad}}_y=\left[\begin{array}{ccc}-3& -10& -3\\ 0& 0& 0\\ +3& +10& +3\end{array}\right]$

Adopt Grad again _xcarry out convolution to image and obtain its vertical edge, it is as follows that its vertical edge extracts formula:

f(x，y)＝f(x，y)*Grad _x

Wherein, f (x, y) is former figure.

Preferably, in step (7), the concrete grammar carrying out local auto-adaptive threshold process is:

(7-1) image is divided into the zonule that m × n etc. is large;

(7-2) carry out OTSU process in each zonule, the two-dimensional matrix that all threshold values form is denoted as Mat_T ₀;

(7-3) to all Mat_T ₀the gaussian filtering carrying out 3 × 3 exports Mat_T ₁, then according to Mat_T ₁carry out the image binaryzation in each region.

Preferably, step (8) is specially:

(8-1) open operation by morphology and noise filtering is carried out to image, specifically first construct a m ₁× n ₁template, then image is first corroded to the operation of rear expansion;

(8-2) make vertical edge in image be fused into connected region by morphology closed operation, specifically first construct a m ₂× n ₂template, the operation of the post-etching that then image first expanded;

Wherein, the edge of object is corroded by corrosion, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) be placed in the center of template, according to template size, go through all over all other pixels covered by template, the value of amendment pixel (x, y) is value minimum in all pixels.Expand then contrary with corrosion, the profile of object is expanded, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) center of template is placed in, according to template size, go through all over all other pixels covered by template, in amendment template, the value of all pixels is value maximum in all pixels.

Preferably, step (9) is specially:

(9-1) zone marker is carried out to connected region, and calculate its minimum enclosed rectangle;

(9-2) according to length breadth ratio ASPECT and the size AREA of Chinese car plate, the deviation ratio e of a given length breadth ratio ₁with the deviation ratio e of area ₂, make the length breadth ratio aspect of the minimum enclosed rectangle of each connected region and area area carry out following inspection:

|aspect-ASPECT|≤e ₁ASPECT

|area-AREA|≤e ₂AREA

The connected region simultaneously meeting above-mentioned condition can carry out step (9-3), otherwise is non-license plate area;

(9-3) in general, the angle of inclination of the car plate in image is not too large, therefore the threshold value T at a given angle of inclination _angle, make the angle of inclination angle of each connected region by following inspection:

|angle-0|≤T _angle

If meet the connected region of above-mentioned condition, think license plate area, otherwise be non-license plate area.

Compared with prior art, tool has the following advantages and beneficial effect in the present invention:

1, the present invention Scharr operator replaces Sobel operator to carry out rim detection, can more adequately inspection vehicle board vertical edge and maintain same high-level efficiency.

2, the present invention introduces intelligibility evaluation module innovatively, and the input picture different to sharpness makes adaptive adjustment, is more conducive to subsequent treatment, the accuracy rate of the larger License Plate that improve under different sharpness.

3, the present invention introduces luminance evaluation module innovatively, makes illumination normalization, be more conducive to subsequent treatment to the input picture of brightness exception, the larger accuracy rate that improve License Plate under complex illumination environment.

Accompanying drawing explanation

Fig. 1 is the general flow chart of a kind of Location Method of Vehicle License Plate based on sharpness and luminance evaluation of the present invention.

Embodiment

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited thereto.

Embodiment

As shown in Figure 1, a kind of Location Method of Vehicle License Plate based on sharpness and luminance evaluation of the present embodiment, specifically comprises the steps:

1, the intelligibility evaluation based on noise and acutance is carried out to input picture:

1. noise evaluation obtains binary map by first carrying out OTSU threshold process to image, then calculates area in connected region and be less than the accounting of given threshold value, obtain noise evaluation value finally by linear transformation.The formula of calculating noise assessed value is as follows:

$\mathrm{Noise}=a\underset{{\mathrm{Area}}_i=0}{\overset{T_{\mathrm{area}}}{\mathrm{\&Sigma;}}}{\mathrm{Comp}}_i/\underset{{\mathrm{Area}}_i=0}{\overset{\&infin;}{\mathrm{\&Sigma;}}}{\mathrm{Comp}}_i+b$

Wherein, Comp _ii-th region, Area _ithe area in i-th region, T _areabe the given threshold value judging that whether area is too small, a and b is the parameter of linear transformation.

What 2. assess sharpness adopted is point sharpness method, gets 8 neighborhood points subtract each other with it often of image, and (size of power depends on distance, distance, then weights are little, and the difference needs as 45 ° and 135 ° directions are taken advantage of first to ask the weighted sum of 8 differences ), by have a some income value to be added divided by the total number of pixel.The formula calculating assess sharpness value is as follows:

$\mathrm{Sharpness}=\underset{i=1}{\overset{m\&times;n}{\mathrm{\&Sigma;}}}\underset{a=1}{\overset{8}{\mathrm{\&Sigma;}}}\left|\frac{\mathrm{df}}{\mathrm{dx}}\right|/(m\&times;n)$

Wherein, a is a field pixel, m and n is the length of image and wide, and df is the change amplitude of gray scale, and dx is the distance increment between pixel.

3. intelligibility evaluation is made up of noise evaluation and assess sharpness, after the noise evaluation value obtaining image and assess sharpness value, is obtained the intelligibility evaluation value of image by certain weighted connections.The formula calculating intelligibility evaluation value is as follows:

Quality＝wNoise+(1-w)Sharpness

Wherein, w is weighted value, and span between zero and one.

2, carry out Grads Sharp process to the image of sharpness deficiency, the too high image of sharpness carries out Gaussian Blur process:

1. given two threshold value T _quality1and T _quality2, wherein T _quality1the threshold value judging that whether image definition is not enough, T _quality2it is the threshold value judging that whether image definition is too high.

2. as Quality < T _quality1time Grads Sharp process is carried out to image, namely ask for image gradient information Grad by Laplace operator, be then added with former figure and obtain Grads Sharp result, the formula of Grads Sharp is as follows:

f(x，y)＝f(x，y)+Grad(x，y)

Wherein, f (x, y) for former figure, Grad (x, y) be gradient information.

3. as Quality > T _quality2time Gaussian Blur process is carried out to image, namely by using Gaussian function to calculate in the Fuzzy Template of normal distribution, and do convolution algorithm with former figure and obtain Gaussian Blur result, the formula of Gaussian function and Gaussian Blur is as follows:

$\mathrm{Gauss}(a,b)=\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}{e}^{-\frac{{(a-m/2)}^2+{(b-n/2)}^2}{2{\mathrm{\&sigma;}}^2}}$

f(x，y)＝f(x，y)*Gauss(a，b)

Wherein, σ is the standard deviation of normal distribution, m and n is that template is long and wide, a and b is element in template, and f (x, y) is former figure.

3, be converted to gray-scale map by RGB figure, conversion formula is as follows:

Gray(x，y)＝0.299R(x，y)+0.587G(x，y)+0.114B(x，y)

Gray(x，y)＝(38R(x，y)+75G(x，y)+15B(x，y))＞＞7

$\mathrm{Gray}(x,y)={(0.2973R{(x,y)}^{2.2}+0.6274G{(x,y)}^{2.2}+0.0753B{(x,y)}^{2.2})}^{\frac{1}{2.2}}$

Wherein, R, G, B are the value of red, green, blue three Color Channels of pixel (x, y), and Gray is the gray-scale value of pixel (x, y) correspondence.

4, luminance evaluation is carried out to gray-scale map:

Carry out the too high or luminance shortage two kinds of situations of brightness to gray level image to assess, specifically calculate pixel value lower than threshold value T _bright1or higher than threshold value T _bright2pixel account for the ratio sum of all pixels, if be greater than given threshold value T _bright3then thinking, is that brightness is abnormal.

If 5 brightness are abnormal, carry out illumination normalization:

1. first correct with Gamma and strengthen brightness of image, Gamma updating formula is as follows:

g(x，y)＝c[f(x，y)] ^γ

In formula, f (x, y) is input picture, and g (x, y) is output image, c be arbitrary value for adjusting picture contrast, γ between 0 ~ 1 for adjusting brightness of image;

2. then carry out difference of Gaussian filtering (DOG, Differnce of Gaussian) and remove illumination high frequency components, i.e. uneven part, DoG Filtering Formula is as follows:

g(x，y)＝f(x，y)*Gauss(σ ₂)-f(x，y)*Gauss(σ ₁)

In formula, f (x, y) is input picture, and g (x, y) is output image, Gauss (σ _n) for variance be σ _nthe Gaussian function of (n=1,2);

3. finally adopt contrast equalization to strengthen the contrast of image irradiation, thus at utmost retain the useful details of image and strengthen picture contrast while of reaching uneven in removal illumination, contrast equalization formula is as follows:

$g_1(x,y)=\frac{f(x,y)}{{\left(\mathrm{mean}\left({\left|f(x^{\&prime;},y^{\&prime;})\right|}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_2(x,y)=\frac{g_1(x,y)}{{\left(\mathrm{mean}\left(\min {(\mathrm{\&tau;},|g_1(x^{\&prime;},y^{\&prime;})\left|\right)}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_3(x,y)=\tanh \left(\frac{g_2(x,y)}{\mathrm{\&tau;}}\right)$

In formula, f (x, y) is input picture, g _n(x, y) (n=1,2,3) are output images, and α is the impact of compressibility index for reducing large pixel value, and τ amputates large pixel value in image after being used for first step normalization;

6, by vertical edge in Scharr operator extraction image:

In order to extract image edge information, conventional has Sobel operator, but because Sobel operator calculates often accurately little on the convolution mask of 3 × 3, therefore the Sobel operator having to improve is called Scharr operator, its operation efficiency is equally fast with Sobel operator, but result is more accurate, and Scharr template is expressed as follows:

${\mathrm{Grad}}_x=\left[\begin{array}{ccc}-3& 0& +3\\ -10& 0& +10\\ -3& 0& +3\end{array}\right]{\mathrm{Grad}}_y=\left[\begin{array}{ccc}-3& -10& -3\\ 0& 0& 0\\ +3& +10& +3\end{array}\right]$

Because horizontal edge is generally harmful to without profit on the contrary for License Plate, only adopt Grad here _xcarry out convolution to image and obtain its vertical edge, it is as follows that its vertical edge extracts formula:

f(x，y)＝f(x，y)*Grad _x

Wherein, f (x, y) is former figure.

7, local auto-adaptive threshold process is carried out:

1. image is divided into the zonule that m × n etc. is large;

2. carry out OTSU process in each zonule, the two-dimensional matrix that all threshold values form is denoted as Mat_T ₀;

3. to all Mat_T ₀the gaussian filtering carrying out 3 × 3 exports Mat_T ₁, then according to Mat_T ₁carry out the image binaryzation in each region;

8, by Morphological scale-space filtered noise, and vertical edge is made to be fused into connected region:

1. open operation by morphology and noise filtering is carried out to image, specifically first construct a m ₁× n ₁template, then image is first corroded to the operation of rear expansion;

2. make vertical edge in image be fused into connected region by morphology closed operation, specifically first construct a m ₂× n ₂template, the operation of the post-etching that then image first expanded;

Wherein, the edge of object is corroded by corrosion, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) be placed in the center of template, according to template size, go through all over all other pixels covered by template, the value of amendment pixel (x, y) is value minimum in all pixels.Expand then contrary with corrosion, the profile of object is expanded, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) center of template is placed in, according to template size, go through all over all other pixels covered by template, in amendment template, the value of all pixels is value maximum in all pixels.

9, zone marker is carried out to connected region, carries out screening and obtaining license plate area according to the feature of Chinese car plate:

1. zone marker is carried out to connected region, and calculate its minimum enclosed rectangle;

2. according to length breadth ratio ASPECT and the size AREA of Chinese car plate, the deviation ratio e of a given length breadth ratio ₁with the deviation ratio e of area ₂, make the length breadth ratio aspect of the minimum enclosed rectangle of each connected region and area area carry out following inspection:

|aspect-ASPECT|≤e ₁ASPECT

|area-AREA|≤e ₂AREA

3. the connected region simultaneously meeting above-mentioned condition can carry out step, otherwise is non-license plate area.

3. in general, the angle of inclination of the car plate in image is not too large, therefore the threshold value T at a given angle of inclination _angle, make the angle of inclination angle of each connected region by following inspection:

|angle-0|≤T _angle

If meet the connected region of above-mentioned condition, think license plate area, otherwise be non-license plate area.

Above-described embodiment is the present invention's preferably embodiment; but embodiments of the present invention are not restricted to the described embodiments; change, the modification done under other any does not deviate from Spirit Essence of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (10)

1., based on a Location Method of Vehicle License Plate for sharpness and luminance evaluation, it is characterized in that, comprise the steps:

(1) intelligibility evaluation based on noise and acutance is carried out to input picture;

(2) carry out Grads Sharp process to the image of sharpness deficiency, the too high image of sharpness carries out Gaussian Blur process;

(3) by RGB figure, gray-scale map is converted to the image after step (2) process;

(4) luminance evaluation is carried out to gray-scale map, namely too high the or luminance shortage two kinds of situations of brightness are carried out to gray level image and assess;

(5) if brightness is abnormal, illumination normalization is carried out;

(6) by vertical edge in Scharr operator extraction image;

(7) local auto-adaptive threshold process is carried out;

(8) by Morphological scale-space filtered noise, and vertical edge is made to be fused into connected region;

(9) zone marker is carried out to connected region, carry out screening and obtaining license plate area according to the feature of Chinese car plate.

2. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, in step (1), intelligibility evaluation is made up of noise evaluation and assess sharpness, after the noise evaluation value obtaining image and assess sharpness value, obtained the intelligibility evaluation value of image by certain weighted connections, the formula calculating intelligibility evaluation value is as follows:

Quality＝wNoise+(1-w)Sharpness

Wherein, w is weighted value, and span between zero and one;

Described noise evaluation obtains binary map by first carrying out OTSU threshold process to image, and then calculate area in connected region and be less than the accounting of given threshold value, obtain noise evaluation value finally by linear transformation, the formula of calculating noise assessed value is as follows:

$\mathrm{Noise}=a\underset{{\mathrm{Area}}_i=0}{\overset{T_{\mathrm{area}}}{\mathrm{\&Sigma;}}}{\mathrm{Comp}}_i/\underset{{\mathrm{Area}}_i=0}{\overset{\&infin;}{\mathrm{\&Sigma;}}}{\mathrm{Comp}}_i+b$

Wherein, Comp _ii-th region, Area _ithe area in i-th region, T _areabe the given threshold value judging that whether area is too small, a and b is the parameter of linear transformation;

What described assess sharpness adopted is point sharpness method, gets 8 neighborhood points and subtracts each other with it, first ask the weighted sum of 8 differences to often of image, by have an income value to be added divided by the total number of pixel, the formula calculating assess sharpness value is as follows:

$\mathrm{Sharpness}=\underset{i=1}{\overset{m\&times;n}{\mathrm{\&Sigma;}}}\underset{a=1}{\overset{8}{\mathrm{\&Sigma;}}}\left|\frac{\mathrm{df}}{\mathrm{dx}}\right|/(m\&times;n)$

Wherein, a is a field pixel, m and n is the length of image and wide, and df is the change amplitude of gray scale, and dx is the distance increment between pixel.

3. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, step (2) is specially:

(2-1) given two threshold value T _quality1and T _quality2, wherein T _quality1the threshold value judging that whether image definition is not enough, T _quality2it is the threshold value judging that whether image definition is too high;

(2-2) as Quality < T _quality1time Grads Sharp process is carried out to image, namely ask for image gradient information Grad by Laplace operator, be then added with former figure and obtain Grads Sharp result, the formula of Grads Sharp is as follows:

f(x,y)＝f(x,y)+Grad(x,y)

Wherein, f (x, y) for former figure, Grad (x, y) be gradient information;

(2-3) as Quality > T _quality2time Gaussian Blur process is carried out to image, namely by using Gaussian function to calculate in the Fuzzy Template of normal distribution, and do convolution algorithm with former figure and obtain Gaussian Blur result, the formula of Gaussian function and Gaussian Blur is as follows:

$\mathrm{Gauss}(a,b)=\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}{e}^{-\frac{{(a-m/2)}^2+{(b-n/2)}^2}{2{\mathrm{\&sigma;}}^2}}$

f(x,y)＝f(x,y)*Gauss(a,b)

Wherein, σ is the standard deviation of normal distribution, m and n is that template is long and wide, a and b is element in template, and f (x, y) is former figure.

4. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, in step (3), the formula being turned gray-scale map by RGB figure is:

Gray(x,y)＝0.299R(x,y)+0.587G(x,y)+0.114B(x,y)；

Or Gray (x, y)=(38R (x, y)+75G (x, y)+15B (x, y)) >>7;

Or $\mathrm{Gray}(x,y)={(0.2973R{(x,y)}^{2.2}+0.6274G{(x,y)}^{2.2}+0.0753B{(x,y)}^{2.2})}^{\frac{1}{2.2}};$ Wherein, R, G, B are the value of red, green, blue three Color Channels of pixel (x, y), and Gray is the gray-scale value of pixel (x, y) correspondence.

5. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, in step (4), the method for gray scale being carried out to luminance evaluation is:

Calculate pixel value lower than threshold value T _bright1or higher than threshold value T _bright2pixel account for the ratio sum of all pixels, if be greater than given threshold value T _bright3then thinking, is that brightness is abnormal.

6. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, in step (5), the method for illumination normalization is:

(5-1) first correct enhancing brightness of image with Gamma, Gamma updating formula is as follows:

g(x,y)＝c[f(x,y)] ^γ

In formula, f (x, y) is input picture, and g (x, y) is output image, c be arbitrary value for adjusting picture contrast, γ between 0 ~ 1 for adjusting brightness of image;

(5-2) then carry out difference of Gaussian filtering and remove illumination high frequency components, i.e. uneven part, difference of Gaussian Filtering Formula is as follows:

g(x,y)＝f(x,y)*Gauss(σ ₂)-f(x,y)*Gauss(σ ₁)

In formula, f (x, y) is input picture, and g (x, y) is output image, Gauss (σ _n) for variance be σ _nthe Gaussian function of (n=1,2);

(5-3) finally adopt contrast equalization to strengthen the contrast of image irradiation, thus at utmost retain the useful details of image and strengthen picture contrast while of reaching uneven in removal illumination, contrast equalization formula is as follows:

$g_1(x,y)=\frac{f(x,y)}{{\left(\mathrm{mean}\left({\left|f(x^{\&prime;},y^{\&prime;})\right|}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_2(x,y)=\frac{g_1(x,y)}{{\left(\mathrm{mean}\left(\min {(\mathrm{\&tau;},|g_1(x^{\&prime;},y^{\&prime;})\left|\right)}^{\mathrm{\&alpha;}}\right)\right)}^{\frac{1}{\mathrm{\&alpha;}}}}$

$g_3(x,y)=\tanh \left(\frac{g_2(x,y)}{\mathrm{\&tau;}}\right)$

In formula, f (x, y) is input picture, g _n(x, y) (n=1,2,3) are output images, and α is the impact of compressibility index for reducing large pixel value, and τ amputates large pixel value in image after being used for first step normalization.

7. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, is characterized in that, in step (6), by the concrete grammar of vertical edge in Scharr operator extraction image be:

Scharr template is expressed as follows:

${\mathrm{Grad}}_x=\left[\begin{array}{ccc}-3& 0& +3\\ -10& 0& +10\\ -3& 0& +3\end{array}\right]{\mathrm{Grad}}_y=\left[\begin{array}{ccc}-3& -10& -3\\ 0& 0& 0\\ +3& +10& +3\end{array}\right]$

Adopt Grad again _xcarry out convolution to image and obtain its vertical edge, it is as follows that its vertical edge extracts formula:

f(x,y)＝f(x,y)*Grad _x

Wherein, f (x, y) is former figure.

8. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, in step (7), the concrete grammar carrying out local auto-adaptive threshold process is:

(7-1) image is divided into the zonule that m × n etc. is large;

(7-2) carry out OTSU process in each zonule, the two-dimensional matrix that all threshold values form is denoted as Mat_T ₀;

(7-3) to all Mat_T ₀the gaussian filtering carrying out 3 × 3 exports Mat_T ₁, then according to Mat_T ₁carry out the image binaryzation in each region.

9. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, step (8) is specially:

(8-1) open operation by morphology and noise filtering is carried out to image, specifically first construct a m ₁× n ₁template, then image is first corroded to the operation of rear expansion;

(8-2) make vertical edge in image be fused into connected region by morphology closed operation, specifically first construct a m ₂× n ₂template, the operation of the post-etching that then image first expanded;

Wherein, the edge of object is corroded by corrosion, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) center of template is placed in, according to template size, go through all over all other pixels covered by template, amendment pixel (x, y) value is value minimum in all pixels, expand then contrary with corrosion, the profile of object is expanded, detailed process is to each pixel (x in image by template, y) following process is done: by pixel (x, y) center of template is placed in, according to template size, go through all over all other pixels covered by template, in amendment template, the value of all pixels is value maximum in all pixels.

10. the Location Method of Vehicle License Plate based on sharpness and luminance evaluation according to claim 1, it is characterized in that, step (9) is specially:

(9-1) zone marker is carried out to connected region, and calculate its minimum enclosed rectangle;

(9-2) according to length breadth ratio ASPECT and the size AREA of Chinese car plate, the deviation ratio e of a given length breadth ratio ₁with the deviation ratio e of area ₂, make the length breadth ratio aspect of the minimum enclosed rectangle of each connected region and area area carry out following inspection:

|aspect-ASPECT|≤e ₁ASPECT

|area-AREA|≤e ₂AREA

The connected region simultaneously meeting above-mentioned condition can carry out step (9-3), otherwise is non-license plate area;

(9-3) the threshold value T at a given angle of inclination _angle, make the angle of inclination angle of each connected region by following inspection:

|angle-0|≤T _angle

If meet the connected region of above-mentioned condition, think license plate area, otherwise be non-license plate area.