机械本科毕业论文外文资料翻译-基于人工神经网络的车牌照识别.doc

中国地质大学长城学院本科毕业论文外文资料翻译系 别： 工程技术系 专 业： 机械设计制造及其自动化 姓 名： 学 号： 2012 年 3 月 20 日外文资料翻译译文基于人工神经网络的车牌照识别厄尔丁克kocera，kursatcevikbK摘要近年来，随着车辆数量在交通中的增加必要的个人工作在交通控制中的数量也随之增加。为了解决这个问题，计算机自动控制系统被开发。汽车牌照自动识别系统就是其中之一。在这个系统中，汽车牌照自动识别系统是基于人工神经网络的。在这个系统中，259个车辆图片被使用。这些车辆的图片是从相机中提取，然后车牌区域尺寸220x50的像素决定了这张照片使用的图像处理算法。字符包括字母和数字，在车牌定位中使用边缘检测算子和斑点的着色方法。斑点染色方法应用于ROL来区分车牌特征。在这一阶段的工作特征提取，采用平均绝对偏差公式。数字化特征进行分类使用前馈多层感知器神经网络回传播。关键词：车辆牌照识别，人工神经网络，模糊着色，字符识别1.简介在发展中国家，汽车数量日益增加。与此同时，必须认识到车辆和车牌同时也是增加的。以计算机为基础的车辆牌照自动识别系统为解决这一问题提供了必要性。在这项研究中，提出了一种高效的汽车牌照自动识别系统基于人工神经网络（神经网络）。该系统由三个主要议题：定位板地区的汽车图像，车牌字符图像的分割，字符分割和识别。该方案提出的车牌自动识别系统显示在图1。CameraVehicleFinding theSegmenting theFeatureRecognitionImagePlate RegionCharactersExtractionwith ANN图1 汽车牌照自动识别系统2.以前的作品根据土耳其民用车牌识别，及成功率（SR）为基础的车牌定位（PRL），字符分割（CS）和字符识别（CR）过程给出了表1 表1AuthorYearNumber of Image UsedSR for PRL (%)SR for CS (%)SR for CR (%)H. Caner20064292,8587,1794,12S. Ozbay200634097,6596,1898,82G. Yavuz200880929590B. Yalim200820096-92,53.定位车牌区域第一阶段的汽车牌照识别系统是找到车牌定位车辆图像。板区域通常由白底黑字组成。因此，过渡区之间的黑色和白色的颜色是非常密集的，在这一调查区域，包括大部分的过渡点，将足已定位车牌区域。为此，边缘检测算子应用于车辆的图像得到的过渡点。坎尼边缘检测器使用了一个过滤器的基础上的一阶导数的高斯平滑。经过平滑的形象，消除噪音，下一步就是提取图像的梯度。这一进程， 33矩阵被作为操作使用尺寸来进行边缘强度的梯度计算。这一信息使我们得到边缘点，如此密集的地区可确定过渡点。过渡点之间的黑色和白色的颜色确定了这一边缘图。边缘检测和定位车牌区域的图像显示在图2图2（一）原始汽车图像边缘检测；（二）局部区域4.字符分割灰度车牌图像分割过程之前应加强。因为对比度差异可能发生在提取图像的照相机。此外，不必要的肮脏的地区和一些噪音影响可以放在分割过程中负方向。在这项工作中，灰度图像的增强了运用对比的延伸和中值滤波技术。因此，对比差异图像和声音等脏区域在白色背景，该板可以消除。图像增强阶段后，斑点显色法的实施，确定边界的字符。4.1.对比度扩展扩展的图像对比度的手段，均衡直方图。换句话说，对比度扩展使图像锐化。灰度直方图是图像灰度分布值的图像。直方图均衡化是一种流行的技术，以改善外观形象差的对比。这个过程直方图均衡化的图像有4个步骤：（1）求直方图的值。（2）规范这些值除以总像素。（3）乘以这些正常价值的最大灰度值。（4）图的新的灰度值。对比度扩展车牌图像显示在图3。4.2.中值滤波中值滤波是用来消除不必要的噪音的地区。在这个滤波算法中周围的图像的33矩阵被截取。这个矩阵的尺寸可以根据噪声水平来进行调整。这个过程的工作，（1）一个像素为中心像素的33矩阵，（2）周围像素邻域像素分配，（3）排序过程之间采用这九个像素由小做大，（4）第五个元素分配为中位数元，（5）这些程序实施的所有像素图像。过滤后的图像显示在图3。图3（一）原车牌区域的图像；（二）对比度扩展图像；（三）中值滤波后的图像4.3.斑点的着色方法斑点（二进制大对象）着色算法具有很强的结构计算法来确定临近和相关地区二进制图像。该算法使用一种特殊形模板扫描，图像从左到右，从上到下。这种扫描过程确定独立的地区获得连接到四方向从零开始的背景图像。在这项工作中，四个方向的点着色算法应用于二进制编码的车牌图像获取字。实施后，字符分割得到了车牌区域的图像（图4）。图4字符分割在这项工作中，字符分割被列为单独的数字和字母。为此，板图像分为三个地区。第一区域包括双位数字，表明城市交通代码。二区域由一至三的字母。第三个区域由二至四位数字组成。板图像扫描形成确定自左向右水平和空间之间的字符。在这个过程，如果该值的空间是高于以前所鉴定的阈值。数字被定位为28x35像素大小。字母被定位为30x 40像素大小。样本的一些数字和字母分割的车牌区域显示在图5。图5一些样品的字符分割5.特征提取在这项研究中，获得的字符被保存为一个图像文件。数字的尺寸被确定为28x35像素，字母的尺寸被确定为30x 40像素。数字和字母进行单独的神经网络传输以提高识别的成功率。在分类之前，图像的特征应该被准确提取。特征提取能使我们获得直观的图像信息。这些信息可以作为一个特征向量。特征向量是包括全球和地方特点的一个字符编码，比较字符就可以确定特征。在拟议的方法，特征向量的虹膜图像编码使用平均绝对偏差算法。该算法的定义是：V1f (x, y) m(1)NNN是在图像的像素数，m是指图像的平均值，f（，y）是在点（，你）的值。在这项工作中，数字图像分为45像素尺寸的图像和字母图像分为55像素尺寸的图像。每个子图像进行特征提取采用AAD。我们得到特征向量长度49字节的数字和48字节的字母。整个特征向量应用到神经网络的输入进行分类提取特征。6.识别特征在我们的工作中，数字和字母进行单独的神经网络传输以提高识别的成功率。他们都有相同的结构，但只有输入数量差异。之所以使用不同的神经网络识别是防止类似的数字和字母被混淆，比如“0”和”O“啊”，“2”“Z”和“8”“B”。我们可以知道，这种复杂性将减少识别成功率。在拟议的方法，多层感知器模型是用于特征分类的。该处理单元被安排在MLP的多层感知器。这些输入层（包括信息，您可以使用作出决定），隐层（帮助网络计算更复杂的关系）和输出层（包括由此产生的决定）。每一个神经元的输入层是直接反馈到隐层神经元，在隐层，综合和产品的重量和隐层神经元输出计算每个输出层节点。如果错误计算输出值与期望值大于错误率，那么培训（改变重量和计算新的输出使用新的重量）过程开始。这个训练过程可以获得理想的错误率。训练神经网络，前馈反向传播算法选择。均方误差（均方差）的功能是用来测量培训网络性能的。值的均方差是用来确定如何适合网络输出所需的输出。标准监督训练通常是基于均方误差。培训计划终止时，均方误差下降到阈值。均方误差值接近零，计算输出值成为接近所需的输出值。7.实验结果为了评估系统的效能，259车辆图像被应用。快速BP学习算法用于训练神经网络。最大的5000次分别为每个输入设置。当系统达到最小错误率，由用户定义，迭代将停止。定义的最小错误率为该应用程序0001。只有一个输入图像是用于测试的系统，其余的是在训练阶段。iteration-mse图形的最佳结果为每个字符数据显示在图6。培训达到最小的误差率在4457个迭代的数量和1180个迭代的信件。图6均方误差迭代图形的训练过程成功率的车牌区域定位，字符分割和字符识别阶段，该系统在表2中给出了。因此，247个车牌图像被正确识别，所以整体识别率的系统是95,36%。表2。成功率的汽车牌照自动识别系统StageNumber of SamplesNumber of Correct ResultsSuccess Rate (%)PRL25925598,45CS25525298,82CR347 (Letters) + 1022 (Numbers)344 (Letters) + 1000 (Numbers)98,17外文原文Procedia Computer Science 3 (2011) 10331037/locate/procediaWCIT-2010Artificial neural networks based vehicle license plate recognitionH.Erdinc Kocera, K.Kursat CevikbaSelcuk Uni. Technical Education Faculty, Konya 42250, TurkeybNigde Uni. Bor Vocational High School, Nigde 51700, TurkeyAbstractIn recent years, the necessity of personal working in traffic control is increasing because the numbers of vehicles in traffic is increasing. To deal with this problem, computer based automatic control systems are being developed. One of these systems is automatic vehicle license plate recognition system. In this work, the automatic vehicle license plate recognition system based on artificial neural networks is presented. In this system, 259 vehicle pictures were used. These vehicle pictures were taken from the CCD camera and then the license plate region dimensioned by 220x50 pixels is determined from this picture by using image processing algorithms. The characters including letters and numbers placing in the license plate were located and determined by using Canny edge detection operator and the blob coloring method. The blob coloring method was applied to the ROI for separation of the characters. In the last phase of this work, the character features were extracted by using average absolute deviation formula. The digitized characters were then classified by using feed forward back propagated multi layered perceptron neural networks. The correct classification rates were given in last section.c 2010 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Guest Editor.Keywords: Vehicle licence plate recognition, artificial neural networks, blob coloring, character recognition.1. IntroductionWithin the fast developing countries, the number of vehicles is increasing day by day. In parallel to this, the need to recognize the vehicles and their license plates is increasing. To supply this necessity, computer based automatic vehicle license plate recognition systems are being developed recently. In this study, we proposed an efficient automatic vehicle license plate recognition system based on artificial neural networks (ANN). This system consists of three major topics. These are; localizing the plate region from the car image, segmenting the characters from the license plate image and recognizing the segmented characters. The block scheme of the proposed automatic license plate recognition system is shown in Fig 1.CameraVehicleFinding theSegmenting theFeatureRecognitionImagePlate RegionCharactersExtractionwith ANNFig. 1. The block scheme of an automatic license plate recognition systemThe layout of this work can be analyzed into 7 section. In section 1, the general information about this work was introduced, the previous works on license plate recognition was given as table in section 2. In section 3, localization1877-0509 c 2010 Published by Elsevier Ltd. doi:10.1016/cs.2010.12.1691034H. Erdinc Kocer, K. Kursat Cevik / Procedia Computer Science 3 (2011) 10331037H.Erdinc Kocer / Procedia Computer Science 00 (2010) 000000process was described. The segmentation process of the characters was introduced in section 4. In section 5, the feature extraction process of the segmented characters were presented. The recognition of the characters by using ANN was presented in Section 6. The experimental results were presented in the last section of this work.2. Previous WorksThe previous works will be held according to the works on Turkish civil license plate recognition. The success rates (SR) for the stage of plate region localization (PRL), character segmentation (CS) and character recognition (CR) processes were given in Table 1 1-6.Table 1. Success rates of the previous worksAuthorYearNumber of Image UsedSR for PRL (%)SR for CS (%)SR for CR (%)H. Caner20064292,8587,1794,12S. Ozbay200634097,6596,1898,82G. Yavuz200880929590B. Yalim200820096-92,5I. Irmakci200814596,5596,6195,25K. Bora2009225 (plate)-10089,333. Localization of the Plate RegionThe first stage of license plate recognition system is finding the plate location from vehicle image. The plate region consists of white background and black characters normally. Therefore, the transitions between black and white colors is very intensive in this region. Finding the region that includes most transition points would be adequate for localizing the plate region.For this purpose, Canny edge detection operator was applied to the vehicle images to get the transition points. The Canny edge detector uses a filter based on the first derivative of a Gaussian smoothing. After smoothing the image and eliminating the noise, the next step is to find the edge strength by taking the gradient of the image. For this process, this operator uses 3x3 dimensioned matrices. The edge strength of the gradient is then calculated. This information gives us the edge points, so the intensive transition points region can be determined. The transition points between black and white colors was then determined from this edge map. The original, the edge detected and the localized plate region images were shown in Fig 2a, 2b and 2c respectively.Fig. 2. (a) original car image; (b) edge detected; (c) localized plate region4. Segmentation of the CharactersThe gray level license plate image should be enhanced before segmentation process. Because the contrast differences can be occurred while taking the images by camera. Also, unwanted dirty regions can be placed on the plate and these noises affects the segmentation process in negative direction.In this work, the gray level plate images were enhanced by applying contrast extension and median filtering techniques. So, the contrast differences between images and the noises such as dirty regions in white background ofH. Erdinc Kocer, K. Kursat Cevik / Procedia Computer Science 3 (2011) 103310371035H.Erdinc Kocer/ Procedia Computer Science 00 (2010) 000000the plate can be eliminated. After image enhancement phase, the blob coloring method was implemented to determine the boundaries of the characters.4.1. Contrast extensionTo extend the contrast of an image means equalization of the histogram of that image. In other words, the contrast extension makes the image sharpen. The gray-level histogram of an image is the distribution of the gray level values in an image. The histogram equalization is a popular technique to improve the appearance of a poor contrasted image.The process of equalizing the histogram of an image consists of 4 steps 7: (1) Find the sum of the histogram values. (2) Normalize these values dividing by the total number of pixels. (3) Multiply these normalized values by the maximum gray-level value. (4) Map the new gray level values. The contrast extended license plate image is shown in Fig 3b.4.2. Median filteringMedian filter is used for eliminating the unwanted noisy regions. In this filtering method, the 3x3 matrices is passed around the image. The dimension of this matrices can be adjusted according to the noise level.The process is working as 7; (1) one pixel is chosen as center pixel of the 3x3 matrices, (2) the arounding pixels are assigned as neigborhood pixels, (3) the sorting process are employed between these nine pixels from smaller to the bigger, (4) the fifth element is assigned as median element, (5) these procedures are implemented to the all pixels in plate image. The filtered plate image is shown in Fig 3c.Fig. 3. (a) original plate region image; (b) contrast extended image; (c) median filtered image4.3. The blob coloring methodThe blob (Binary Large Object) coloring algorithm has a strong architecture to determine the closed and contacless regions in a binary image. This algorithm uses a special L shaped template to scan the image from left to right and from up to down. This scanning process determine the independent regions by obtaining the connections into four direction from zero valued background. In this work, four directional blob coloring algorithm is applied to the binary coding license plate image for getting the characters 8. After implementation, the segmented characters were obtained from the license plate region image (Fig 4).Fig. 4. The segmented charactersIn this work, the segmented characters were classified as numbers and letters separately. For this purpose, the plate image was divided into three region. The first region consists of two digit numbers that indicates the city traffic code. The second region consists of one-to-three digit letters. The third region consists of two-to-four digit numbers. The plate image was scanned form left to right horizontally and the spaces between characters were determined in this process. If the value of the space is higher than threshold value than the character region is signed. The numbers were localized as 28x35 pixels dimension. The letters were localized as 30x40 pixels dimension. Some samples of numbers and letters segmented from plate region are shown in Fig 5.Fig. 5. Some samples of the segmented characters1036H. Erdinc Kocer, K. Kursat Cevik / Procedia Computer Science 3 (2011) 1033103H.Erdinc Kocer / Procedia Computer Science 00 (2010) 000005. Feature ExtractionIn this study, the obtained characters were saved as an image file separately. The dimension of the numbers was determined as 28x35 pixels, the dimension of the letters was determined as 30x40 pixels. The numbers and the letters were classified by using two separate ANN for increasing the success rate of the recognition phase. Before classification, the character images should be feature extracted. Feature extraction provides us to obtain the most discriminating information of an image. This information can be presented as a feature vector. A feature vector that includes global and local features of an character should be encoded so that the comparison between characters can be made. In the proposed approach, the feature vector of an iris image was encoded by using Average Absolute Deviation algorithm 9. This algorithm is defined as:V1f (x, y) m(1)NNwhere N is the number of pixels in the image, m is the mean of the image and f(x,y) is the value at point (x,y). In this work, the number images were divided into 4x5 pixels dimensioned sub-images and the letter images were divided into 5x5 pixels dimensioned sub-images. Each sub-image were feature extracted by applying AAD. We obtained the feature vectors with the length of 49 byte for numbers and 48 byte for letters. The entire feature vectors were applied to the ANN as an input for classification of the characters.6. Recognition of the CharactersIn our work, the numbers and the letters were classified by using two separate ANN for increasing the success rate of the recognition phase. Both of them have same architecture but only the input numbers were differed. The reason for using two separate ANN for recognition is preventing the complexity of recognition of similar numbers and letters such as “0” “O”, “2” “Z” and “8” “B”. As we can know, this complexity will decrease the recognition success.In the proposed approach, a multi layered perceptron (MLP) ANN model was used for classification of the characters. The processing units in MLP are arranged in three layers. These are input layer (includes the information you would use to make decision), hidden layer (helps network to compute more complicated associations) and output layer (includes the resulting decision) 10,11. Each neuron in the input layer is fed directly to the hidden layer neurons via a series of weights. The sum of the products of the weights and the inputs is calculated in each node. The calculated values are fed directly to the output layer neurons via a series of weights. As in hidden layer, the sum of the products of the weights and the hidden layer neuron outputs is calculated in each node in the output layer. If the error between calculated output value and the desired value is more than the error ratio, then the training (changing the weights and calculating the new output by using the new weights) process begins. This training process can be finished by obtaining the desired error rate for all input combinations.For training the ANN, feed-forward back-propagation algorithm was chosen. For measuring the training performance of the network, mean square error (MSE) function is used. The value of the MSE is used to determine how well the network output fits the desired output. The stop criteria for supervised training are usually based on MSE. Most often the training is set to terminate when the MSE drops to some threshold. Approaching the MSE value to the zero means that the calculated output value is becoming the closer to the desired output value.7. Experimental ResultsIn order to evaluate the performance of the proposed system, 259 vehicle images were employed. Sigmoidfunction is used in the activation of neurons. Quick back propagation learning algorithm was used for training the ANN. Maximum 5000 iterations were performed for each input set. When the system reach to the minimum error rate which defined by the user, the iterations will be stopped. The defined minimum error rate for this application is 0,001. Only one of the input images was used for testing the system, the rest was performed in training phase. The iteration-MSE graphics of best results for each character data set are shown in Fig 6. The training reaches the minimum error rate in 4457 iterati