License plate image background noise reduction method based on gradient descent critical point

Abstract

The invention provides a gradient descent critical point-based license plate image background noise reduction method, which is based on a gradient descent thought and finds a direction in which a function value descends fastest through a step-by-step iteration mode, so that a dynamic license plate image noisy point threshold value can be calculated and noise reduction is carried out, the noise in the image can be smoothed, and the noise in the image can be reduced. Meanwhile, detail information of the image can be reserved as much as possible. According to the method, the unfixed threshold can be calculated according to the image, and the method has the advantages of being easy to implement, simple in calculation and wide in application range.

Description

Technical Field

[0001] The invention belongs to the field of intelligent transportation, and in particular is a method for reducing background noise of a license plate image based on a gradient descent critical point. Background Art

[0002] The application of dynamic license plate image background noise reduction method can optimize the image according to the environment, improve the adaptability of the license plate recognition system in different scenarios, enhance the stability and reliability of license plate recognition, and enable the license plate recognition system to still work efficiently in complex environments. It provides solid support for intelligent transportation systems, parking lot management, violation monitoring, etc., and promotes the development of society in a smarter and safer direction.

[0003] At present, dynamic license plate image background noise reduction mainly adopts a variety of methods, which cover traditional image processing technology and modern deep learning algorithms. Commonly used methods include statistical filtering method, wavelet transform method, deep learning algorithm, etc.

[0004] Traditional methods such as mean, median and Gaussian filtering remove noise by statistically analyzing pixel neighborhood values, but they can easily lead to blurred image edges and loss of details; wavelet transform uses multi-scale decomposition to separate noise and signal in the frequency domain, effectively retaining image features, but the computational complexity is high and parameter adjustment is difficult; deep learning, especially convolutional neural networks (CNNs), learns noise patterns through large amounts of data training, has strong denoising capabilities and good generalization, and is applicable to a variety of noise and image content, but has high requirements for data quality and scale, consumes large amounts of computing resources, and has complex internal mechanisms in the model, making it less interpretable. At present, these methods have their own advantages, but they also have limitations and the results are not ideal. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a license plate image background denoising method based on gradient descent critical point. In image processing, gradient is usually used to describe the spatial rate of change of image pixel values, thereby helping to identify image features such as edges and textures.

[0006] The technical solution of the present invention is as follows:

[0007] A method for background noise reduction of license plate images based on gradient descent critical points comprises the following steps:

[0008] 1) Import the image into MATLAB and store it as the matrix img1 variable. Since the original image is a color 3D image, extract one dimension through img1(:,:,1) and store it as img2 variable for subsequent analysis;

[0009] 2) Import the obtained matrix into Excel, set the value of step length Len, and iterate and calculate

[0010] 3) Calculate the maximum number of loop steps:

[0011] SumImg = zeros(MaxIterN, 1)

[0012] CurrentIter=1

[0013] While CurrentIter <MaxIterN

[0014] for i=1:m

[0015] for j=1:n

[0016] If img ij -CurrentIter*Len<0

[0017] img ij =0

[0018] else

[0019] img ij =img ij -CurrentIter*Len

[0020] end

[0021] end

[0022] Sumimg CurrentIter =sum(img)

[0023] CurrentIter=CurrentIter+1

[0024] end

[0025] 4) Calculate the slope of the sequence Sumimg

[0026]

[0027] 5) Determine the current image threshold based on the gradient descent critical point

[0028] ThreLoc=Peak(SlopeSumImg)

[0029] 6) Determine the background noise threshold.

[0030] Thre=ThreLoc*Len

[0031] 7) Import the data after threshold subtraction into MATLAB for analysis to obtain the license plate image after background noise reduction.

[0032] In one embodiment of the present invention, the size of the picture in step 1) is 35 x 119 pixels.

[0033] In one embodiment of the present invention, in step 1), the img2 variable is 35 rows and 119 columns.

[0034] In one embodiment of the present invention, in step 2), the step length is set to: Len=3.

[0035] The beneficial effects of the present invention are as follows: the background noise reduction method of the license plate image based on the gradient descent critical point of the present invention is based on the idea of ​​gradient descent, and finds the direction in which the function value decreases fastest by step-by-step iteration, thereby calculating the noise threshold of the dynamic license plate image and performing noise reduction, which can not only smooth the noise in the image, but also retain the detailed information of the image as much as possible. The method can calculate a non-fixed threshold value according to the image itself, and has the advantages of easy implementation, simple calculation and wide application range. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the original image matrix of the present invention;

[0037] Figure 2 It is the Sumimg iterative change line chart of the present invention;

[0038] Figure 3 It is the Sumimg broken line slope change graph of the present invention;

[0039] Figure 4 is the license plate image before noise reduction of the present invention;

[0040] Figure 5 This is the license plate image after noise reduction of the present invention. DETAILED DESCRIPTION

[0041] Example

[0042] In this embodiment, a method for background noise reduction of a license plate image based on a gradient descent critical point of the present invention comprises the following steps:

[0043] 1) Place the image ( Figure 4 ) (size is 35 x 119 pixels) and imported into MATLAB and stored as the matrix img1 variable. Since the original image is a color 3D image, one of the dimensions is extracted through img1 (:,:,1) and stored as the img2 variable (35 rows and 119 columns) for subsequent analysis.

[0044] 2) If Figure 1 , import the obtained matrix into Excel, set the step length: Len = 3, and calculate the result of each iteration

[0045] 3) Calculate the maximum number of loop steps:

[0046] SumImg = zeros(MaxIterN, 1)

[0047] CurrentIter=1

[0048] While CurrentIter <MaxIterN

[0049] for i=1:m

[0050] for j=1:n

[0051] If img ij -CurrentIter*Len<0

[0052] img ij =0

[0053] else

[0054] img ij =img ij -CurrentIter*Len

[0055] end

[0056] end

[0057] Sumimg CurrentIter =sum(img)

[0058] CurrentIr=CurrentIter+1

[0059] end( Figure 2 )

[0060] 4) Calculate the slope of the sequence Sumimg

[0061]

[0062] 5) Determine the current image threshold based on the gradient descent critical point

[0063] ThreLoc=Peak(SlopeSumImg)

[0064] 6) Determine the background noise threshold.

[0065] Thre=ThreLoc*Len

[0066] 7) Import the data after subtracting the threshold into MATLAB for analysis, and you can get the license plate image after background noise reduction ( Figure 5 ).

[0067] Table 1 Parameters used in the present invention:

[0068]

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims

1. A background denoising method for license plate images based on gradient descent critical points, characterized in that The steps include: 1) Import the image into MATLAB and store it as the matrix img1 variable. Since the original image is a color 3D image, extract one dimension through img1(:,:,1) and store it as img2 variable for subsequent analysis; 2) Import the obtained matrix into Excel, set the value of step length Len, and iterate and calculate 3) Calculate the maximum number of cycle steps; 4) Calculate the slope of the sequence Sumimg; 5) Determine the current image threshold according to the gradient descent critical point; 6) Determine the background noise threshold; 7) Import the data after threshold subtraction into MATLAB for analysis to obtain the license plate image after background noise reduction.

2. The method according to claim 1, characterized in that Step 3) The method for calculating the maximum number of cycle steps is: SumImg = zeros(MaxIterN, 1) CurrentIter=1 While CurrentIter <MaxIterN for i=1:m for j=1:n If img ij -CurrentIter*Len<0 img ij =0 else img ij =img ij -CurrentIter*Len end end Sumimg CurrentIter =sum(img) CurrentIter=CurrentIter+1 end 3. The method according to claim 1, characterized in that Step 4) The method for calculating the slope of the sequence Sumimg is:

4. The method according to claim 1, characterized in that Step 5) The method for determining the current image threshold according to the gradient descent critical point is: ThreLoc=Peak(SlopeSumImg).

5. The method according to claim 1, characterized in that Step 6) The method for determining the background noise threshold is: Thre = ThreLoc*Len.

6. The method according to claim 1, characterized in that The size of the image in step 1) is 35 x 119 pixels.

7. The method according to claim 1, characterized in that In step 1), the img2 variable has 35 rows and 119 columns.

8. The method according to claim 1, characterized in that In step 2), the step length is set to: Len=3.

Images