CN110674821B - A non-motor vehicle license plate recognition method - Google Patents

Abstract

Translated fromChinese

本发明公开了一种非机动车车牌识别方法，本发明先利用BP神经网络对对非机动车原车牌图像之中的数字进行数字识别，获取数字识别后的非机动车原车牌之中的数字区域。其次对所要识别的目标区域的名称进行搜索和扩展，对扩展后的完整目标区域作为样本输入，并构建Faster R‑CNN网络。然后进行目标区域检测提取。最后对提取到的非机动车目标区域，利用BP神经网络进行字符、数字的识别。本发明不仅保持了深度学习技术在目标检测识别上的优势，有效地增强了对非机动车车牌的目标区域检测的精准度以及检测识别的速度，还提升了BP神经网络对非机动车车牌之中的字符、数字的识别率。

The invention discloses a non-motor vehicle license plate recognition method. The invention first uses a BP neural network to perform digital recognition on the numbers in the original non-motor vehicle license plate image, and obtains the numbers in the digitally recognized original non-motor vehicle license plate. area. Secondly, search and expand the name of the target area to be identified, take the expanded complete target area as a sample input, and build a Faster R-CNN network. Then perform target region detection and extraction. Finally, for the extracted non-motor vehicle target area, the BP neural network is used to identify characters and numbers. The invention not only maintains the advantages of the deep learning technology in target detection and recognition, but also effectively enhances the detection accuracy of the target area of the non-motor vehicle license plate and the speed of detection and recognition, and also improves the BP neural network to the non-motor vehicle license plate. The recognition rate of characters and numbers in .

Description

Translated fromChinese

一种非机动车车牌识别方法A non-motor vehicle license plate recognition method

技术领域technical field

本发明涉及一种目标检测识别的方法，具体是基于一种目标区域提取结合Faster-RCNN以及BP神经网络识别非机动车车牌字符、数字的方法。The invention relates to a method for target detection and identification, in particular a method for identifying characters and numbers of non-motor vehicle license plates based on target area extraction combined with Faster-RCNN and BP neural network.

背景技术Background technique

非机动车车牌与机动车车牌不同，机动车车牌的格式一般都是蓝底矩形，其上的字符为一个中文字符(一般为省市级的简称)加上一个英文字母，再加上其后五位字符(一般都是由字母和数字所组成)，其车牌除车牌号字符以外的前后左右各位置均无干扰，格式也较为统一。但是，非机动车车牌的格式一般由该市下辖的区名以及一连串数字组成，且该串数字长度不等，譬如某些颜色的非机动车车牌还存在不带市区号的现象，某些颜色的车牌区号用“临时”的字样代替，格式严重不统一，可谓五花八门，且在车牌的顶部位置，往往会出现“XX市”、“XX市交通局”、“电动自行车”等小字字样，若利用传统方法直接对车牌图像进行字符识别，在实践中发现，会出现如下错误：The non-motor vehicle license plate is different from the motor vehicle license plate. The format of the motor vehicle license plate is generally a rectangle with a blue bottom. With five characters (generally composed of letters and numbers), the license plate has no interference in the front, back, left, and right positions except for the license plate number characters, and the format is relatively uniform. However, the format of the non-motor vehicle license plate is generally composed of the district name under the city and a series of numbers, and the length of the series of numbers varies. The area code of the license plate is replaced by the word "temporary". The format is seriously inconsistent, and it can be described as various, and at the top of the license plate, there are often small words such as "XX City", "XX City Transportation Bureau", "Electric Bicycle", etc. If the traditional method is used to directly perform character recognition on the license plate image, it is found in practice that the following errors will occur:

①扫描字符时提示越界出错。①When scanning characters, it prompts an out-of-bounds error.

②扫描字符时误将上面的干扰字符作为车牌字符，识别出错。② When scanning characters, the above interference characters are mistakenly regarded as license plate characters, and the recognition is wrong.

③由于干扰字符太多，导致识别的车牌字符(包括数字)准确度降低。③ Due to too many interfering characters, the accuracy of the recognized license plate characters (including numbers) is reduced.

发明内容SUMMARY OF THE INVENTION

针对现有技术的不足，本发明提供了一种目标区域提取结合Faster-RCNN以及BP神经网络先检测提取非机动车车牌的目标区域、再在目标区域上识别非机动车车牌的字符、数字的方法。In view of the deficiencies of the prior art, the present invention provides a target area extraction method combining Faster-RCNN and BP neural network to first detect and extract the target area of the non-motor vehicle license plate, and then identify the characters and numbers of the non-motor vehicle license plate on the target area. method.

本发明先利用特定方法获取到非机动车车牌中所要识别的目标区域，并将其作为样本以便后续训练Faster R-CNN网络，也就是将整个非机动车车牌图像看成一个原始图像，将要识别的非机动车车牌目标区域看成原始图像中所要检测的目标。即先利用BP神经网络识别非机动车车牌中的数字字符，以此获得非机动车车牌中的数字区域，再在该数字区域的基础之上利用边界扩展法，扩展至目标区域；再将该目标区域作为输入样本构建Faster R-CNN网络并在以后的检测中以该网络来检测并获取非机动车车牌的目标区域，然后再利用BP神经网络识别该车牌中的目标区域中的字符、数字。The invention first uses a specific method to obtain the target area to be identified in the non-motor vehicle license plate, and uses it as a sample for subsequent training of the Faster R-CNN network, that is, the entire non-motor vehicle license plate image is regarded as an original image, which will be recognized The target area of the non-motor vehicle license plate is regarded as the target to be detected in the original image. That is, first use BP neural network to identify the digital characters in the non-motor vehicle license plate, so as to obtain the digital area in the non-motor vehicle license plate, and then use the boundary expansion method on the basis of the digital area to expand to the target area; The target area is used as the input sample to construct the Faster R-CNN network, and the network is used to detect and obtain the target area of the non-motor vehicle license plate in the subsequent detection, and then use the BP neural network to identify the characters and numbers in the target area in the license plate. .

步骤1、利用BP神经网络对对非机动车原车牌图像之中的数字进行数字识别。Step 1. Use the BP neural network to digitally recognize the numbers in the original license plate image of the non-motor vehicle.

步骤2、获取数字识别后的非机动车原车牌之中的数字区域。Step 2: Obtain the digital area in the original license plate of the non-motor vehicle after the digital identification.

步骤3、对非机动车原车牌图像所要识别的目标区域的名称进行搜索和扩展。Step 3: Search and expand the name of the target area to be recognized by the original license plate image of the non-motor vehicle.

步骤4、对扩展后的完整目标区域作为样本输入，并构建Faster R-CNN网络。Step 4. Use the expanded complete target area as a sample input, and construct a Faster R-CNN network.

步骤5、利用构建的网络对非机动车原车牌图像进行目标区域检测提取。Step 5. Use the constructed network to detect and extract the target area on the original license plate image of the non-motor vehicle.

步骤6、对提取到的非机动车目标区域，利用BP神经网络进行字符、数字的识别。Step 6: Use BP neural network to identify characters and numbers in the extracted non-motor vehicle target area.

本发明将BP神经网络、目标区域截取方法与Faster R-CNN网络方法相结合，不仅保持了深度学习技术在目标检测识别上的优势，有效地增强了对非机动车车牌的目标区域检测的精准度以及检测识别的速度，还提升了BP神经网络对非机动车车牌之中的字符、数字的识别率。The invention combines BP neural network, target area interception method and Faster R-CNN network method, which not only maintains the advantages of deep learning technology in target detection and recognition, but also effectively enhances the accuracy of target area detection of non-motor vehicle license plates. It also improves the recognition rate of characters and numbers in non-motor vehicle license plates by BP neural network.

本发明将问题转化为利用Faster R-CNN以及其所带的边框拟合功能对原始的非机动车车牌中的特定目标的检测，实践证明，利用该种方法，可以有效“过滤”非机动车原车牌图像的上下、左右存在的干扰字符、并且提高BP神经网络对非机动车车牌图像之中的字符、数字的识别率。The invention transforms the problem into the detection of specific targets in the original non-motor vehicle license plate by using Faster R-CNN and the frame fitting function it brings. Practice has proved that this method can effectively "filter" non-motor vehicles. Interfering characters in the upper, lower, left and right sides of the original license plate image, and improve the recognition rate of characters and numbers in the non-motor vehicle license plate image by BP neural network.

附图说明Description of drawings

图1为该方法的步骤流程总图；Fig. 1 is the general flow chart of the steps of the method;

图2(a)为非机动车原始车牌图像的数字识别图；Figure 2(a) is a digital recognition diagram of the original license plate image of a non-motor vehicle;

图2(b)为图2(a)识别所得的数字区域；Figure 2(b) is the digital area identified by Figure 2(a);

图3为非机动车车牌底色为黄时的原图；Figure 3 is the original picture when the background color of the non-motor vehicle license plate is yellow;

图4(a)为非机动车车牌底色为黄时所截得的数字区域；Figure 4(a) is the digital area intercepted when the background color of the non-motor vehicle license plate is yellow;

图4(b)为非机动车车牌底色为黄时所扩展的非机动车车牌的目标区域；Figure 4(b) is the expanded target area of the non-motor vehicle license plate when the background color of the non-motor vehicle license plate is yellow;

图5为非机动车车牌底色为白时的原图；Figure 5 is the original picture when the background color of the non-motor vehicle license plate is white;

图6为非机动车车牌底色为白时所截得的数字区域；Figure 6 is the digital area intercepted when the background color of the non-motor vehicle license plate is white;

图7为非机动车车牌底色为白时所扩展的非机动车车牌的目标区域。Figure 7 shows the expanded target area of the non-motor vehicle license plate when the background color of the non-motor vehicle license plate is white.

具体实施方式Detailed ways

以下结合附图对本发明作进一步说明，如图1所示，本发明具体包括以下步骤：The present invention will be further described below in conjunction with the accompanying drawings. As shown in Figure 1, the present invention specifically includes the following steps:

步骤1：对于所获得的非机动车原车牌图像，判断其底色，若为白，则车牌底色不变，若为黄，则分别将车牌底色转换为白色，并调用它们的图像直方图，设置背景像素的像素均值为N₁，字符像素的像素均值为N₂，其中(N₁>N₂)，对于底色为其他颜色的非机动车车牌，若有需要，亦可以将其底色变换为白色，将其字符、数字变换为黑色，此处以底色为白、字符、数字为黑和底色为黄、字符、数字为黑的非机动车车牌为例。Step 1: For the obtained original license plate image of non-motor vehicles, determine its background color. If it is white, the background color of the license plate will remain unchanged. If it is yellow, then convert the background color of the license plate to white, and call their image histograms. Figure, set the pixel mean value of background pixels to N₁ , and the pixel mean value of character pixels to N₂ , where (N₁ >N₂ ), for non-motor vehicle license plates whose background colors are other colors, if necessary, you can also use The background color is changed to white, and its characters and numbers are changed to black. Here, a non-motor vehicle license plate with white background, black characters and numbers, and yellow background and black characters and numbers is used as an example.

步骤2：识别该非机动车原车牌图像之中的数字，并获得数字区域，见图2(a)和图2(b)。Step 2: Identify the numbers in the original license plate image of the non-motor vehicle, and obtain the number area, as shown in Figure 2(a) and Figure 2(b).

步骤3：第一次判断位数，若识别的非机动车车牌的底色为白色且所获取的数字区域之中的数字位数大于6位，则丢弃(见图5，若车牌底色为白色时，其上的数字至多为6位，若大于6位，即可判定所获取的非机动车原车牌图像之中的数字区域是错误的)；若识别的非机动车车牌的底色为黄色且所获取的数字区域之中的数字位数大于7位，则丢弃(见图3，若车牌底色为黄色时，其上的数字至多为7位，若大于7位，即可判定所获取的非机动车原车牌图像之中的数字区域是错误的)。Step 3: Judging the number of digits for the first time, if the background color of the identified non-motor vehicle license plate is white and the number of digits in the obtained digital area is greater than 6 digits, discard it (see Figure 5, if the background color of the license plate is When it is white, the number on it is at most 6 digits. If it is greater than 6 digits, it can be determined that the number area in the obtained original license plate image of non-motor vehicle is wrong); if the background color of the recognized license plate of non-motor vehicle is Yellow and the number of digits in the obtained digital area is greater than 7, discard it (see Figure 3, if the background color of the license plate is yellow, the number on it is at most 7, if it is greater than 7, it can be determined that the The digital area in the acquired non-motor vehicle original license plate image is wrong).

步骤4：第二次判断位数，若识别的非机动车车牌的底色为白色且所获取的数字区域之中的数字位数等于6位，则保留，转到步骤11；若识别的非机动车车牌的底色为黄色且所获取的数字区域之中的数字位数等于7位，则转到步骤10。Step 4: Determine the number of digits for the second time. If the background color of the recognized non-motor vehicle license plate is white and the number of digits in the obtained digital area is equal to 6 digits, keep it and go to step 11; If the background color of the motor vehicle license plate is yellow and the number of digits in the obtained digital area is equal to 7 digits, then go to step 10 .

步骤5：第三次判断位数，若识别的非机动车车牌的底色为白色且所获取的数字区域之中的数字位数小于6位，则转到步骤7；若识别的非机动车车牌的底色为黄色且所获取的数字区域之中的数字位数小于7位，则转到步骤7。Step 5: Judging the number of digits for the third time, if the background color of the recognized non-motor vehicle license plate is white and the number of digits in the obtained digital area is less than 6, then go to step 7; if the recognized non-motor vehicle license plate is If the background color of the license plate is yellow and the number of digits in the obtained number area is less than 7, go to step 7.

步骤6：对步骤5中位数有所缺失的车牌，分别先做直方图均衡化、灰度化、边缘检测、以及形态学处理，先从该串缺失数字串的最左侧开始搜索其第一位数字的位置，以白底黑字(即白底非机动车车牌)的车牌为例，首先判断竖直方向上的每一列，若除自己本身以外，一个像素点的周边像素点的像素值σ满足Step 6: Perform histogram equalization, grayscale, edge detection, and morphological processing on the license plates with missing medians in Step 5, and start searching for the number plate from the leftmost part of the missing number string. The position of one digit, taking a license plate with black characters on a white background (that is, a non-motor vehicle license plate on a white background) as an example, first determine each column in the vertical direction, if, except for itself, the surrounding pixels of a pixel are pixels The value σ is satisfied

|σ-N₂|<|σ-N₁| (I)|σ-N₂ |<|σ-N₁ | (I)

的像素个数超过5个，则认为该像素在数字字符之上或数字字符旁边，若该像素所在列的像素值小于200的像素个数超过10个时，判断该像素点已经落在数字字符之上，字符的水平方向也基于此类方法判断。If the number of pixels in the column exceeds 5, it is considered that the pixel is above or next to the numeric character. If the number of pixels in the column where the pixel value is less than 200 exceeds 10, it is judged that the pixel has fallen on the numeric character. Above, the horizontal direction of characters is also judged based on such methods.

步骤7：以该像素点为中心点，分别向左、向右进行搜索，直到左边和右边都出现像素值小于200的像素且不满足步骤6中所述条件时，判定这两个像素为该数字字符的边界像素，设这两个像素之间的每个像素点的距离为1，以此计算这两个像素的距离，设为该字符的最大宽度d_x。Step 7: Take the pixel as the center point, and search to the left and right respectively, until there are pixels whose pixel value is less than 200 on the left and right, and when the conditions described in step 6 are not met, determine that these two pixels are the The boundary pixel of the digital character, set the distance of each pixel between the two pixels to 1, and calculate the distance between the two pixels, set it as the maximum width d_x of the character.

步骤8：按照步骤6、步骤7的方法迭代搜索，直到对每一个数字字符都得出了它们的最大宽度

(其中i表示步骤5中所得的数字字符的位数)，并求它们的平均值，设为

Step 8: Iteratively search according to steps 6 and 7 until each digit character has its maximum width

(where i represents the number of digits of the numeric characters obtained in step 5), and find their average value, set as

步骤9：设一个经验值为θ，阈值为T，此时从步骤5中所得的第一个数字字符的最左侧和最后一个字符的最右侧开始，按照步骤6、步骤7的方法搜索，若获得的字符最大宽度满足式(II)，则可以判断为另一个数字，扩展边界；Step 9: Set an empirical value of θ and a threshold value of T. At this time, starting from the leftmost of the first digital character and the rightmost of the last character obtained in step 5, search according to the methods of steps 6 and 7. , if the maximum width of the obtained character satisfies the formula (II), it can be judged as another number to expand the boundary;

步骤10：若非机动车车牌的底色为黄色且所获取的数字区域中之中的数字位数正确，则将该图像作为非机动车原车牌之中的目标区域输出；若非机动车车牌的底色为白色且所获取的数字区域中之中的数字位数正确，则继续进行步骤11；否则继续执行步骤5—步骤9。Step 10: If the background color of the non-motor vehicle license plate is yellow and the number of digits in the obtained digital area is correct, the image is output as the target area in the original non-motor vehicle license plate; If the color is white and the number of digits in the acquired digital area is correct, proceed to step 11; otherwise, proceed to step 5-step 9.

步骤11：继续按最左边第一位数字的最左侧进行搜索并对该数字字符进行区域标记求和，记为∑x₁，若搜索到有像素值满足(I)式，则继续在垂直方向和水平方向按照步骤6的方法进行搜索，若该像素周边的像素值满足(I)式的像素个数小于3个，则按照区域搜索的方法对这些像素进行标记求和，记为∑x₂，若

则判断该区域为中文字符与数字字符之间的分隔点，否则，以该像素点为中心，进行水平方向的向右搜索，直到出现水平方向上的某一个像素点的八邻域内的像素点的像素值均步满足(I)式为止，以该像素值所在列为分割线，分割车牌，并扩展边界至整个车牌图像的左侧边界，作为最终车牌目标区域，见图4(b)、、图6、图7。Step 11: Continue to search by the leftmost of the first digit on the leftmost and carry out the area mark summation for the numerical character, denoted as ∑ x₁ , if there is a pixel value that meets the formula (I), continue to The direction and the horizontal direction are searched according to the method of step 6. If the number of pixels around the pixel that satisfies the formula (I) is less than 3, these pixels are marked and summed according to the method of area search, denoted as ∑x₂ , if

Then it is judged that the area is the separation point between Chinese characters and numeric characters, otherwise, take the pixel as the center, and perform a rightward search in the horizontal direction until a pixel in the eight neighborhoods of a certain pixel in the horizontal direction appears. Until the pixel value of the pixel value satisfies the formula (I), take the pixel value as the dividing line, segment the license plate, and extend the boundary to the left edge of the entire license plate image, as the final license plate target area, see Figure 4(b), , Figure 6, Figure 7.

步骤12：将步骤10、步骤11所得的非机动车车牌目标区域在原车牌图像中计算比例、边界角等相关参数，并作为输入Faster R-CNN网络的样本图像，构建的Faster R-CNN网络。Step 12: Calculate the scale, boundary angle and other related parameters of the non-motor vehicle license plate target area obtained in step 10 and step 11 in the original license plate image, and use it as the sample image input to the Faster R-CNN network to construct the Faster R-CNN network.

步骤13：再利用构建的Faster R-CNN网络检测并获取非机动车原车牌图像的目标区域。Step 13: Reuse the constructed Faster R-CNN network to detect and obtain the target area of the original license plate image of non-motor vehicles.

步骤14：对步骤13中检测所得的非机动车原车车牌图像的目标区域，利用BP神经网络进行非机动车车牌的字符、数字识别。Step 14: For the target area of the original non-motor vehicle license plate image detected in step 13, use the BP neural network to perform character and number recognition of the non-motor vehicle license plate.

说明：其中步骤1-步骤11阶段为样本采集阶段，包括数字区域的获得、目标区域的边界扩展、车牌数字位数的判别等；其中步骤12-步骤13阶段为构建网络训练阶段，包括Faster R-CNN网络的构建、利用构建的网络检测非机动车车牌原图像目标区域等；步骤14阶段为非机动车原始车牌图像目标区域字符识别阶段，包括对非机动车车牌原始图像目标区域进行字符、数字识别等。Description: The steps 1-11 are the sample collection stages, including the acquisition of the digital area, the boundary expansion of the target area, the discrimination of the number of digits of the license plate, etc.; the steps 12-13 are the stages of network construction and training, including Faster R -Construction of CNN network, use the constructed network to detect the target area of the original image of the non-motor vehicle license plate, etc.; Step 14 is the character recognition stage of the target area of the original image of the non-motor vehicle license plate, including character recognition of the target area of the original image of the non-motor vehicle license plate. digital recognition, etc.

Claims (1)

1. A method for recognizing a license plate of a non-motor vehicle is characterized in that:

the whole license plate image is regarded as an original image, a non-motor vehicle license plate area to be identified is regarded as a target to be detected in the original image, and a BP neural network is utilized to identify digital characters in the non-motor vehicle license plate so as to obtain a digital area in the non-motor vehicle license plate;

expanding to a target area by using a boundary expansion method on the basis of the digital area; taking the target area as an input sample, constructing a Faster R-CNN network, and detecting the target area of the license plate of the non-motor vehicle by using the network in the subsequent detection;

recognizing characters and numbers in a target area in the license plate by using a BP neural network;

the original image needs to be preprocessed, and specifically comprises the following steps: judging the background color of the obtained non-motor vehicle original image, and if the non-motor vehicle original image is white, keeping the background color of the license plate unchanged; if the background color is yellow, respectively converting the bottom color of the license plate into white, converting the characters and the numbers into black, calling image histograms of the characters and the numbers, and setting the pixel mean value of the background pixels to be N₁The mean value of the pixels of the character is N₂，N₁>N₂；

The expanding to the target area by using the boundary expanding method specifically comprises the following steps:

step 1: judging the digit for the first time, and if the identified license plate is a white-bottom license plate and the digit is more than 6 digits, discarding the license plate; if the identified license plate is a yellow bottom license plate and the number digit is more than 7 digits, discarding the license plate;

step 2: judging the digit for the second time, if the recognized license plate is a white-bottom license plate and the digit is equal to 6 digits, reserving the recognized license plate, and turning to the step 9; if the identified license plate is a yellow bottom license plate and the number digit is equal to 7 digits, turning to the step 8;

and step 3: judging the digit for the third time, and if the recognized license plate is a white-bottom license plate and the digit is less than 6, turning to the step 4; if the identified license plate is a yellow bottom license plate and the number digit is less than 7 digits, turning to the step 4;

and 4, step 4: respectively carrying out histogram equalization, graying, edge detection and morphological processing on the license plate with the missing median in the step 3, firstly searching the position of the first digit of the missing digit string from the leftmost side of the string, taking the license plate with a white-black character as an example, firstly judging each column in the vertical direction, if the number of pixels, of which the pixel value sigma of the peripheral pixels of one pixel meets the formula (I), exceeds 5,

|σ-N₂|<|σ-N₁| (I)

if the number of the pixels of which the pixel value is less than 200 exceeds 10, judging that the pixel point falls on the digital character, and judging the horizontal direction of the character based on the mode;

and 5: respectively searching leftwards and rightwards by taking the pixel point as a central point until the pixels with the pixel values smaller than 200 appear on the left side and the right side and do not satisfy the formula (I) in the step 4, judging that the two pixels are boundary pixels of the digital character, setting the distance of each pixel point between the two pixels as 1, calculating the distance of the two pixels by using the distance, and setting the distance as the maximum width d of the character_x；

Step 6: and iterating the search in the way of step 4 and step 5 until the maximum width of each digital character is obtained

Wherein i represents the number of digits of the numeric character obtained in step 3, and the average value of the digits is set as

And 7: setting an empirical value as theta and a threshold value as T, searching in the modes of step 4 and step 5 from the leftmost side of the first numeric character obtained in step 3 and the rightmost side of the last character, and if the maximum width of the obtained character meets the formula (II), judging that the character is another number and expanding the boundary;

and 8: if the ground color of the license plate of the non-motor vehicle is yellow and the number digits in the acquired number region are correct, outputting the image as a target region in the original license plate of the non-motor vehicle; if the ground color of the license plate of the non-motor vehicle is white and the digit number in the acquired digit area is correct, continuing to perform the step 9; otherwise, continuing to execute the step 3 to the step 7;

and step 9: continuing to search the leftmost digit of the leftmost digit and performing a region-marked summation of the digit characters, denoted as ∑ x₁If the pixel value satisfies the formula (I), the searching is continued in the vertical direction and the horizontal direction according to the mode of the step 4, if the number of the pixels of which the pixel values around the pixel satisfy the formula (I) is less than 3, the pixels are marked and summed according to the mode of area searching, and the sum is recorded as sigma x₂If, if

Judging that the region is a separation point between the Chinese character and the numeric character; otherwise, the pixel point is taken as the center, the right search in the horizontal direction is carried out until the pixel values of the pixel points in the eight adjacent domains of a certain pixel point in the horizontal direction all meet the formula (I), the column where the pixel value is located is taken as a dividing line, the license plate is divided, and the boundary is expanded to the left side boundary of the whole license plate image to be used as the final license plate target area.

Images