CN106778742B - Car logo detection method based on Gabor filter background texture suppression - Google Patents

Abstract

The invention discloses a vehicle logo detection method based on Gabor filter background texture suppression, comprising the following steps: the first step is to perform tilt correction preprocessing on an image; Obtain the license plate area; the third step, based on the prior knowledge, according to the positional relationship between the license plate and the car logo, after the license plate is located, the rough positioning area of the car logo containing the car logo pattern is obtained; the fourth step, the rough positioning area of the car logo is subjected to Gabor filtering , suppress the texture of the heat dissipation network around the car logo and highlight the car logo area; the fifth step, perform Gaussian filtering and mathematical morphological closing operations; the sixth step, select a threshold to threshold the gray image, and frame the detection target area to realize the Precise positioning. The vehicle logo detection method has short detection time and high detection rate.

Description

A vehicle logo detection method based on Gabor filter background texture suppression

technical field

The invention relates to a vehicle logo detection method, in particular to a vehicle logo detection method based on Gabor filter background texture suppression.

Background technique

With the rapid growth of the social economy, the current demand for automobiles in China is growing, and the number of automobiles is increasing, which also brings traffic problems such as hit-and-run accidents and vehicle theft. In order to identify illegal and illegal vehicles, it is generally used to recognize the license plate of the car. However, in recent years, the phenomenon of license plate reversal, license plate wear, license plate occlusion, etc., makes it unreliable to identify the car only by recognizing the license plate. The car logo is a key image that contains the information of the model and manufacturer, and is an important basis for car classification and identification. If the car logo can be located accurately, it will effectively improve the accuracy of car classification and recognition.

However, car logos are rich in types and shapes, and do not have stable external features. At the same time, the environment in which they are located is the area of the car grille with complex textures. In addition, the car logo is easily affected by the weather: in the case of insufficient light at night or rainy days, the car logo is difficult to identify; under strong light conditions, the car logo is easily reflective. The above characteristics make the positioning of the vehicle logo more difficult and challenging. The existing vehicle logo positioning methods are not mature enough, and have problems such as low detection rate and easy interference from light changes, so the effect of vehicle logo detection needs to be further improved.

SUMMARY OF THE INVENTION

Purpose of the invention: The present invention provides a vehicle logo detection method based on Gabor filter background texture suppression, aiming at the problems existing in the prior art.

Technical solution: The vehicle logo detection method based on Gabor filter background texture suppression according to the present invention includes:

(1) For the vehicle image with a certain tilt angle obtained by shooting, the vehicle symmetry axis detection and tilt correction are performed based on the SIFT operator;

(2) Use the machine learning algorithm of Harr+AdaBoost to train a cascade classifier, and use the cascade classifier to locate the license plate area from the corrected vehicle image;

(3) Based on the prior knowledge, according to the positional relationship between the license plate and the car logo, obtain the rough positioning area of the car logo including the car logo pattern in the located license plate area;

(4) Gabor filtering is performed on the rough positioning area of the car logo to suppress the texture of the heat dissipation network around the car logo and highlight the car logo area;

(5) Perform Gaussian filtering and mathematical morphological closing operations on the vehicle logo area;

(6) Selecting a threshold pair Thresholding the grayscale image obtained in step (5), and framing the detection target area to obtain a precisely positioned vehicle logo.

Beneficial effect: Compared with the prior art, the present invention has the following significant advantages:

1) High detection accuracy: This method has a certain anti-interference to illumination changes, and has a high detection rate under different illumination conditions. Light) have better positioning effect;

2) Good real-time performance: the method of the present invention has a faster detection speed at night or under the condition of insufficient illumination, under the condition of direct sunlight and under the condition of uniform illumination, and for the picture data captured by the highway bayonet. , which can realize online real-time processing;

3) Wide range of applicable objects: The traditional background texture suppression algorithm often judges the texture direction of the rough positioning area of the car logo first, and then adopts different car logo positioning algorithms according to different texture directions. In the wrong case, the wrong positioning method will be used, which will inevitably lead to the failure of positioning, and the method based on Gabor filter background texture suppression adopted in the present invention has high applicability without the need for texture direction discrimination, and can be applied to vehicles at the same time. Vehicles with horizontal, vertical and mesh textured backgrounds are used for vehicle logo detection, so the present invention is an algorithm that is suitable for different situations and has better robustness.

Description of drawings

Fig. 1 is the schematic flow chart of the vehicle logo detection method based on Gabor filter background texture suppression of the present invention;

Fig. 2 is the flow chart of license plate positioning;

FIG. 3 is a schematic diagram of the filtering flow of the Gabor filter.

Detailed ways

As shown in FIG. 1 , the vehicle logo detection method based on Gabor filter background texture suppression in this embodiment includes:

(1) For the vehicle image with a certain tilt angle obtained by shooting, the vehicle symmetry axis detection and tilt correction are performed based on the SIFT operator.

This step specifically includes:

x_i,y_i表示该点坐标，

表示方向，s_i表示尺度信息，i＝1,…,n，n为特征点的总数；(1-1) Obtain the point vector of each feature point of the vehicle image, wherein the point vector defining the feature point i is

x _i , y _i represent the coordinates of the point,

Represents the direction, s _i represents the scale information, i=1,...,n, n is the total number of feature points;

经过归一化后，得到对应的特征描述子k_i，i＝1,…,n，并将所有点的特征描述子生成预设维数的SIFT特征点向量；(1-2) Direction of feature points

After normalization, the corresponding feature descriptors k _i are obtained, i=1,...,n, and the feature descriptors of all points are generated into SIFT feature point vectors of preset dimensions;

(1-3) Generate a mirror image m _i by directly modifying the feature descriptor _ki , and then generate a possible symmetrical feature point pair (pi , p _j ) by matching the feature point and the mirror image, i, _j =1,...,n, i≠j;

(1-4) Calculate the angle confidence Φ _ij , the scale confidence S _ij and the distance confidence D _ij of each pair of possible symmetrical feature point pairs (pi , p _j ), and then calculate the total confidence M _{i , j} , i, j=1,...,n, i≠j; where:

其中，

分别为点p_i和点p_j的方向,θ_ij为点p_i到点p_j的方向；The calculation formula of the angle confidence Φ _ij is:

in,

are the directions of point p _i and point p _j respectively, and θ _ij is the direction from point p _i to point p _j ;

其中σ_s为尺度因子,是高斯函数包络线沿点p_i到点p_j的方向的标准方差；The scale confidence S _ij is obtained by quantifying the similarities s _i and s _j of the scales in p _i and p _j :

where σ _s is the scale factor, which is the standard deviation of the Gaussian function envelope along the direction from point p _i to point p _j ;

σ_d为距离边界，d为为对称点p_i到点p_j的距离；The distance confidence D _ij is:

σ _d is the distance boundary, and d is the distance from the symmetrical point p _i to the point p _j ;

The total confidence M _i,j is:

(1-5) Calculate the symmetry axis r _{ij of each pair of symmetrical feature points (pi , p j )} : r _ij =x _c cosθ _ij + _{y c} sinθ _ij , where x _c , y _c are the feature point pair respectively (p _i ,p _j ) lengths in the x-axis and y-axis directions, θ _ij is the direction from point p _i to point p _j ;

(1-6) Use linear Hough transform to find the main axis of symmetry, each pair of symmetrical feature points (pi , p _j ) to points (r _ij , θ _ij ) in the Hough space with weights M _{i , j} vote to get the main axis of symmetry;

(1-7) Taking the angle of the main axis of symmetry as the vehicle inclination angle θ;

(1-8) Perform tilt correction on the vehicle image according to the vehicle tilt angle θ, and the corrected image is:

width and height are the width and height of the original image, and width' and height' are the width and height of the corrected image.

(2) Use the machine learning algorithm of Harr+AdaBoost to train a cascade classifier, and use the cascade classifier to locate the license plate area from the corrected vehicle image.

Step (2) specifically includes the following steps:

(2-1) Collect a preset number of positive samples and negative samples, and establish a positive sample library and a negative sample library respectively, where the positive sample refers to the license plate cut out from the high-definition vehicle photos taken at the road checkpoint, and the negative sample is Randomly cropped background samples from non-license plate areas in vehicle photos, including a variety of background environments;

(2-2) Extract the Harr features of all positive and negative samples, and use these Harr features to train several strong classifiers of the AdaBoost algorithm;

Among them, the training method of the strong classifier of the AdaBoost algorithm is:

(2-2-1) Let the training set S={(a ₁ ,b ₁ ),...,(am ,b _m )} contain _m samples, where a _i ∈A(i=1,2, ...,m) represents the training sample, A is the training sample set, b _i ∈ B is the discriminant sign corresponding to a _i , and there is B={1,-1}, for the jth weak The classifier h _j (a _i ) is expressed as

Wherein, F _j (a _i ) represents the value of the j-th Haar feature in the sub-window, δ _j represents the set threshold, and p _j represents the amount that controls the direction of the inequality sign;

(2-2-2) Use the AdaBoost algorithm to train the weak classifier into a strong classifier. The specific steps are as follows:

Initialize sample weights:

Among them, w ₁ (i) represents the initial weight of the ith sample in the first round of training, p represents the total number of positive samples in S, q represents the total number of negative samples in S, p+q=m, m is the total number of samples ;

For t=1,2,...,T (T is the number of iterations), the following loop is performed:

①Weight normalization

Among them, w _t (i) represents the weight of the i-th sample in the t-th round of training, i=1,2,...,m;

②Train out its weak classifier h _j for feature j, and calculate its weighted error ε _j , namely

And select the classifier h _{t min} with the lowest weighted error as the classifier for this cycle;

③ Update the sample weights according to the following formula:

Wherein, when the classification is correct, h _t (a _i )=b _i , e _i =0, and when the classification is wrong, h _t (a _i )≠ _bi , e _i =1;

(2-2-3) Get the final strong classifier, as follows

a为待检窗口，h_t(a)表示在第t轮训练中得到的弱分类器，H(a)的结果为1表示接受，0表示拒绝。in

a is the window to be checked, h _t (a) represents the weak classifier obtained in the t-th round of training, and the result of H(a) is 1 for acceptance and 0 for rejection.

(2-3) Construct license plate detection cascade classifiers in the form of cascade from several strong classifiers obtained by training;

(2-4) As shown in Figure 2, the trained license plate detection cascade classifier is used to detect the license plate area in the vehicle image;

(2-5) Based on the license plate scale feature and the license plate position feature, the detected license plate area is picked up and screened by mistake.

Step (2-5) is used to exclude falsely detected areas after localization.

Among them, the proportion of license plate: the shape of the license plate is a rectangle, the license plate has a fixed number of characters and a fixed-size font, the actual width and height are 44 cm and 14 cm, respectively, and the aspect ratio in the image is basically about 3:1. Set size range limits for width and height;

License plate position characteristics: Since the actual road surveillance camera is triggered by the fixed position ground sensing coil signal and the license plate is installed at the lower part of the entire vehicle, the vertical position of the license plate in the entire image is basically stable (more tends to appear in the image). The lower part), according to the statistical information, the average license plate center position can be obtained, and the candidate area that deviates from the center position is less likely to become a license plate.

(3) By observing the topological structure of the front parts of the vehicle, it can be found that the most easily distinguishable feature of the car logo is its position. The biggest difference between it and other interference areas is that the car logo will not appear in other positions except above the license plate. The position of the interference area is random and uncertain, so the approximate range of the car logo can be roughly located by the position of the license plate; through the test of a large number of vehicle pictures, according to the positional relationship between the license plate and the car logo, the located license plate area contains the car logo pattern. The rough positioning area of the vehicle logo is:

Among them, X ₁ , X ₂ , Y ₁ , and Y ₂ are the left and right and upper and lower boundaries of the approximate range of the vehicle logo obtained by rough positioning, respectively, X _left and X _right are the left and right boundaries of the license plate area, and Y _up is the upper boundary of the license plate area, height is the height of the license plate area, and N is an optional height coefficient, generally taken as N=3.

(4) Gabor filtering is performed on the rough positioning area of the car logo to suppress the texture of the heat dissipation network around the car logo and highlight the car logo area.

As shown in Figure 3, step (4) specifically includes the following steps:

(4-1) Define the two-dimensional Gabor kernel function h(x, y) of the Gabor filter as

为设定的方向；Among them, u ₀ is the center frequency of the function, σ _x and σ _y are the scale factors, and are the standard deviations of the Gaussian function envelope along the x and y axes respectively, exp(2πju ₀ R ₁ ) is the oscillation function, and the real part is the cosine function, the imaginary part is a sine function,

is the set direction;

可以提取图像中不同方向的纹理信息，

的取值范围为

这个范围内的

值可以描述所有的方向，即

与

描述的是同一个方向，同样通过改变Gabor核函数的尺度可以提取图像中不同尺度上的纹理信息，本发明中没有选择多尺度上纹理信息的提取，而是在一个效果良好的特定尺度下，选取了6个不同方向的Gabor滤波器构成滤波器组,方向分别为π/6，π/4，π/3，2π/3，3π/4，5π/6。By changing the direction of the Gabor kernel function

The texture information in different directions in the image can be extracted,

The value range of is

within this range

values can describe all directions, i.e.

and

It describes the same direction, and the texture information on different scales in the image can be extracted by changing the scale of the Gabor kernel function. In the present invention, the extraction of texture information on multiple scales is not selected, but at a specific scale with good effect, Six Gabor filters with different directions are selected to form a filter bank, and the directions are π/6, π/4, π/3, 2π/3, 3π/4, 5π/6.

分别设定为π/6，π/4，π/3，2π/3，3π/4，5π/6，6个方向的Gabor滤波器的尺度参数和带宽参数设置如下：尺度σ_x、σ_y的值均设定为1，带宽Sigma设置为2π，且Sigma＝σu₀，故中心频率u₀为2π；(4-2) Initialize Gabor filter: Construct 6 Gabor filters {h _l (x,y)|l=1,...,6} in different directions respectively to form a filter bank, where the direction

Set as π/6, π/4, π/3, 2π/3, 3π/4, 5π/6, respectively, the scale parameters and bandwidth parameters of the Gabor filter in 6 directions are set as follows: scale σ _x , σ _y The values of are all set to 1, the bandwidth Sigma is set to 2π, and Sigma=σu ₀ , so the center frequency u ₀ is 2π;

(4-3) Convolve the image of the rough positioning area of the vehicle logo obtained in step (3) with 6 two-dimensional Gabor kernel functions h(x, y) in different directions and take the modulo to obtain 6 filtered images G ₁ (x,y), G2 ₍ x,y), G3 ₍ x,y), G4 ₍ x,y), _G5 (x,y), _G6 (x,y), filter bank The final output image is G(x,y) ₌ ( _G1 (x,y)+G2(x,y) ₊ G3(x,y) ₊ G4(x,y)+G5 ₍ x,y) )+G ₆ (x,y))/6, where G _l (x,y)=∫∫f(x ₀ ,y ₀ )h _l (xx ₀ ,yy ₀ )dx ₀ dy ₀ , where, f(x,y) is the image of the rough positioning area of the vehicle logo;

(4-4) Calculate its attenuation coefficient λ _pk for each pixel point pk of the final output image G(x,y) of the filter bank obtained in step (4-3), k=1,...,num,num is the total number of pixels, and the calculation steps are as follows:

1. to obtain the image I _x containing horizontal texture and the image I _y containing vertical texture with Sobel horizontal edge detection method and Sobel vertical edge detection method respectively in the rough positioning area obtained after the license plate positioning;

②Calculate the attenuation coefficient of each pixel point pk as:

(x _pk , y _pk ) are the coordinates of the pixel point pk, α and β are two constant coefficients, in the present invention, α=0.5, β=2; width_pai is the width of the license plate area obtained by positioning; it can be seen from the formula: λ The size of _pk is related to I _x /I _y and x _pk : the closer I _x /I _y is to 0 or infinity (corresponding to the case of horizontal or vertical texture), the smaller λ _pk is. Similarly, the more x _pk deviates from width_pai/ 2 places (the common axis of symmetry of the car logo and the license plate), the smaller the λ _pk , the

(4-5) Multiply each pixel of the wave filter group output image G(x,y) by its corresponding attenuation coefficient λ _pk .

(5) Gaussian filtering and mathematical morphological closing operations are performed on the vehicle logo area.

In this step, the two-dimensional discrete Gaussian filter function is applied to the image obtained in step (4) to filter out noise, and gray-level mathematical morphological closing operation is performed to bridge the discontinuity or fracture of the detection target and eliminate small holes in the target.

(6) Selecting a threshold pair Thresholding the grayscale image obtained in step (5), and framing the detection target area to obtain a precisely positioned vehicle logo.

Step (6) specifically includes the following steps:

即(6-1) Select 1/2 of the average pixel gray value of the image T(x _pk , y _pk ) obtained in step (5) as the threshold

which is

(6-2) Thresholding is performed by thresholding, and the image after thresholding is

(6-3) The detection target area is framed from the thresholded image, and this target area is the vehicle logo area.

Claims (7)

1. A car logo detection method based on Gabor filter background texture suppression is characterized by comprising the following steps:

(1) the method comprises the steps of carrying out vehicle symmetry axis detection and inclination correction on a vehicle image which is shot and acquired and has a certain inclination angle based on an SIFT operator;

(2) training by using a machine learning algorithm of Harr + AdaBoost to obtain a cascade classifier, and positioning a license plate region from the corrected vehicle image by using the cascade classifier;

(3) based on prior knowledge, obtaining a vehicle logo coarse positioning area containing vehicle logo patterns in the positioned license plate area according to the position relation between the license plate and the vehicle logo;

(4) gabor filtering is carried out on the vehicle logo coarse positioning area, radiating network textures around the vehicle logo are inhibited, and the vehicle logo area is highlighted; the method specifically comprises the following steps:

(4-1) defining the two-dimensional Gabor kernel function h (x, y) of the Gabor filter as

Wherein u is₀As a function of center frequency, σ_x、σ_yIs a scale factor, which is the standard deviation of the envelope of the Gaussian function along the directions of the x axis and the y axis, exp (2 pi ju)₀R₁) Is an oscillation function, the real part is a cosine function, the imaginary part is a sine function,

is a set direction;

(4-2) initializing Gabor filter: gabor filters { h) of 6 different directions are respectively constructed_l(x, y) | l ═ 1,. ·,6}, thereby constituting a filter bank in which directions are set forth

The dimension parameters and bandwidth parameters of the Gabor filter set to be pi/6, pi/4, pi/3, 2 pi/3, 3 pi/4, 5 pi/6 and 6 directions respectively are set as follows: dimension σ_x、σ_yAre all set to 1, the bandwidth Sigma is set to 2 pi, and Sigma ═ Sigma u₀Center frequency ofu₀Is 2 pi;

(4-3) respectively carrying out convolution operation on the vehicle logo coarse positioning area image obtained in the step (3) and two-dimensional Gabor kernel functions h (x, y) in 6 different directions and carrying out modulus extraction to obtain 6 filtering images G₁(x,y)，G₂(x,y)，G₃(x,y)，G₄(x,y)，G₅(x,y)，G₆(x, y), the final output image of the filter bank is G (x, y) ═ G₁(x,y)+G₂(x,y)+G₃(x,y)+G₄(x,y)+G₅(x,y)+G₆(x, y))/6, wherein G_l(x,y)＝∫∫f(x₀,y₀)h_l(x-x₀,y-y₀)dx₀dy₀In the formula, f (x, y) is an image of the vehicle logo coarse positioning area;

(4-4) calculating the attenuation coefficient lambda of each pixel point pk of the final output image G (x, y) of the filter bank obtained in the step (4-3)_pkAnd k is 1., num, num is the total number of the pixel points, and the calculation steps are as follows:

① obtaining an image I containing horizontal textures by respectively using a Sobel horizontal edge detection method and a Sobel vertical edge detection method in a coarse positioning area obtained after license plate positioning_xAnd an image I comprising vertical texture_y；

② calculating the attenuation coefficient of each pixel pk as:

(x_pk,y_pk) α and β are two constant coefficients for the coordinates of the pixel point pk, and width _ pai is the width of the license plate region obtained by positioning;

(4-5) multiplying each pixel point of the wave filter group output image G (x, y) by the corresponding attenuation coefficient lambda_pk；

(5) Performing Gaussian filtering and mathematical morphology closing operation on the car logo area;

(6) and (5) selecting a threshold value pair to threshold the gray level image obtained in the step (5), and framing the detection target area to obtain the precisely positioned car logo.

2. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 1, wherein the method comprises the following steps: the step (1) specifically comprises the following steps:

(1-1) acquiring a point vector of each feature point of the vehicle image, wherein the point vector of the feature point i is defined as

x_i,y_iThe coordinates of the point are represented by,

indicates the direction, s_iRepresenting scale information, i is 1, …, n, n is the total number of the feature points;

(1-2) Direction of feature points

After normalization, the corresponding feature descriptor k is obtained_iI is 1, …, n, and generating SIFT feature point vectors with preset dimensions from feature descriptors of all points;

(1-3) by directly modifying the feature descriptor k_iGenerating mirror image m_iAnd generating possible symmetrical feature point pairs (p) by matching the feature points and the mirror images_i,p_j)，i，j＝1,…,n，i≠j；

(1-4) calculating each pair of possible symmetrical feature point pairs (p)_i,p_j) Angle confidence of (phi)_ijScale confidence S_ijAnd distance confidence D_ijThen, the total confidence coefficient M is obtained by calculation_i,jI, j ≠ 1, …, n, i ≠ j; wherein:

angle confidence phi_ijThe calculation formula of (2) is as follows:

wherein,

are respectively a point p_iAnd point p_jDirection of (a), theta_ijIs a point p_iTo point p_jThe direction of (a);

scale confidence S_ijBy quantizing p_iAnd p_jMesoscale similarity s_iAnd s_jTo obtain:

wherein sigma_sIs a scale factor, is a point p along the envelope of the Gaussian function_iTo point p_jThe standard deviation of the direction of (a);

distance confidence D_ijComprises the following steps:

σ_dis a distance boundary, d is a symmetry point p_iTo point p_jThe distance of (d);

total confidence M_i,jComprises the following steps:

(1-5) calculating each pair of symmetrical characteristic point pairs (p)_i,p_j) Axis of symmetry r_ij：r_ij＝x_ccosθ_ij+y_csinθ_ijWherein x is_c，y_cAre respectively a pair of characteristic points (p)_i,p_j) Length in x-and y-directions, θ_ijIs a point p_iTo point p_jThe direction of (a);

(1-6) finding a main symmetry axis by using a linear Hough transform, wherein each pair of symmetric characteristic point pairs (p)_i,p_j) For points (r) in Hough space_ij,θ_ij) By weight M_i,jVoting to obtain a main symmetry axis;

(1-7) taking the angle of the main symmetry axis as the vehicle inclination angle theta;

(1-8) performing inclination correction on the vehicle image according to the vehicle inclination angle theta, wherein the corrected image is as follows:

width and height which are the width and height of the original image, and width' and height which are the width and height of the corrected image.

3. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 1, wherein the method comprises the following steps: the step (2) specifically comprises the following steps:

(2-1) collecting a preset number of positive samples and negative samples, and respectively establishing a positive sample library and a negative sample library, wherein the positive samples refer to license plates cut out from high-definition vehicle photos shot at a road gate, and the negative samples refer to background samples randomly cut out from non-license plate areas in the vehicle photos and contain various background environments;

(2-2) extracting Harr characteristics of all positive and negative samples, and training a plurality of strong classifiers of an AdaBoost algorithm by using the Harr characteristics;

(2-3) constructing a license plate detection cascade classifier by a plurality of strong classifiers obtained by training in a cascade mode;

(2-4) detecting a license plate region in the vehicle image by adopting a trained license plate detection cascade classifier;

and (2-5) false detection and screening are carried out on the detected license plate area according to the license plate proportion characteristic and the license plate position characteristic.

4. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 3, wherein the method comprises the following steps: the training method of the strong classifier of the AdaBoost algorithm in the step (2-2) comprises the following steps:

(2-2-1) setting a training set S { (a)₁,b₁),...,(a_m,b_m) Contains m samples, where a_iE.a represents a training sample, i is 1,2_iE B is a_iCorresponding discrimination indicator, and B ═ 1, -1, for the jth weak classifier h of the ith training sample_j(a_i) Is shown as

Wherein, F_j(a_i) Representing the value of the jth Haar feature in the subwindow, δ_jIndicating a set threshold value, p_jA quantity representing a direction of controlling the unequal sign;

(2-2-2) training the weak classifier into a strong classifier by using an AdaBoost algorithm, and specifically comprising the following steps:

initializing sample weights:

wherein, w₁(i) Representing the initial weight of the ith sample in the first round of training, p representing the total number of positive samples in S, q representing the total number of negative samples in S, wherein p + q is m, and m is the total number of samples;

for T1, 2, T being the number of iterations, the following loop is performed:

① weight normalization

Wherein, w_t(i) Representing the weight of the ith sample in the t round of training, wherein i is 1, 2.

② training weak classifier h of feature j_jCalculating its weighted error ε_jI.e. by

And selects the classifier h with the lowest weighted error_tminAs a classifier for this cycle;

③ the sample weights are updated according to the following formula:

wherein when the classification is correct, h_t(a_i)＝b_i，e_i＝0，When the classification is wrong, h_t(a_i)≠b_i，e_i＝1；

(2-2-3) obtaining the final strong classifier as follows

Wherein

a is the window to be inspected, h_t(a) The weak classifiers obtained in the t-th round of training are shown, and the result of H (a) is 1, which indicates acceptance, and 0, which indicates rejection.

5. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 1, wherein the method comprises the following steps: the step (3) specifically comprises the following steps:

based on prior knowledge, according to the position relation of the license plate and the vehicle logo, a vehicle logo coarse positioning area containing a vehicle logo pattern is obtained in the positioned license plate area, wherein the vehicle logo coarse positioning area is as follows:

wherein, X₁、X₂、Y₁、Y₂Left and right and upper and lower boundaries, X, of rough range of vehicle logo obtained by rough positioning_left、X_rightAre the left and right boundaries of the license plate region, Y_upThe upper boundary of the license plate area, height is the height of the license plate area, and N is a selectable height coefficient.

6. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 1, wherein the method comprises the following steps: the step (5) specifically comprises the following steps:

and (4) filtering noise of the image obtained in the step (4) by using a two-dimensional discrete Gaussian filter function, and performing gray-scale morphological closed operation to close the discontinuity or fracture of the detected target and eliminate fine holes in the target.

7. The method for detecting the car logo based on the Gabor filter background texture suppression according to claim 1, wherein the method comprises the following steps: the step (6) specifically comprises the following steps:

(6-1) selecting the image T (x) obtained in the step (5)_pk,y_pk) 1/2 of the mean value of the pixel gray values as a threshold value

Namely, it is

(6-2) thresholding by using a threshold value, wherein the thresholded image is

And (6-3) framing and detecting a target area from the thresholded image, wherein the target area is a car logo area.

Images