The Recognition System Of Characters On Label Based On Artificial Neural Network

The development and application of computer affect our working and lifegreatly. It provides us with a fashion of inputting character that can transportpicture we get from scanning to text, and improves our working efficiencyobviously. Much information will be expressed by several simple characters, whichcan be accepted and understood conveniently. The date of production, the batchnumber of medicine, and the book number for searching in library, the charactersof vehicle license plate, and the code of the post appear everywhere. Our societystepping into information age we are no longer satisfied with getting informationby our eyes and ears, writing it into paper and process it later. Instead we'd like totake place us with computer to do simple and repetitious work, for instance, puttingcharacters into computer and handling them. The study of characters recognitionsystem has much practical value to people's life, industry, national defence,medical treatment and scientific and technology. Licenses plate will be recognizedby the recognition system of characters on label based on artificial neural network.The recognition system of characters on label based on artificial neural network isstudied in this paper and by which licenses plate will be recognized.The number on the vehicle is increasing sharply all the country the economicsdeveloping rather fast all over the world. The transportation is a hard problem tosettle. Many experts all over the world are studying on ITS in order to improve theefficacy of transportation. The license plate recognition is a key technique aboutITS and one of major topic of the computer vision and pattern recognition in thefield of ITS. The information of the vehicle will be clear when the license plate islearned because it's the unique identification of vehicle.The main content in this paper is using the theories of BP algorithm ofArtificial Neural Networks in pattern recognition to realize the recognition aboutcharacters on the license plate. There are three main parts in this system aboutrecognition of license plate. (1)Module about locate licenses plate. The task ofcomplete in this module is to separate licenses plate from the picture of vehicle thatcontained the licenses plate. This part contains the rough location and the accuratelocation .The rough location is used to find the position of licenses plate promptlyand separate it, while the accurate location is used to adjust the gradient of thevehicle license plate and realize to find the exactly position of the license plate. (2)Module of preprocessing picture. This module is to complete removing the framefrom the picture of the licenses plate, separating characters, standardization and thecharacter feature extracted, which is the base of the next step. (3) The module ofneural network recognition. The module of preprocessing precedes the picture witha series of transposition and then gives the trait of characters. Characters will berecognized and the result will be shown.1,The rough location of licenses plateThe edge detection will be used to realize the rough location of the licensesplate in this paper. For the color palette about bit map with 256 colors is rathercomplicated, which make the algorithm of proceeding picture cannot be used, graypreceded must be done with and the picture binaries must be tread. We will makethe picture of license plate binarization with a threshold and apply edge detection tobinary picture. There is much good arithmetic that has the feature in digital imageprocessing. The calculation of each point in edge detection is only related to thetwo points of its neighbors for goal is to fix the location instead of recognition. Theabsoluteness will be selected if the result obtained from the subtraction of the leftpixel and the right pixel is small than zero, then the result can reflect the binariespicture of edge about perpendicular projection. The picture can be process withthe module in consideration of characters contain many short erect segments andthe background noise has a large part of isolation, the position can be detected bythe characteristic of frequent changing in the area of characters .2,The precise location of licenses plate .The picture of the licenses plate we intercept here may be italic and has frame.We must finish the task of the precise location of the licenses plate namely wemake the vehicle licenses plate horizontal in order to make the recognitionconvenient. The test of angle about the italic vehicle licenses plate can be turned tothe italic angle about the longest segment that we get from Hough transposition,but the result cannot make us satisfied. The product of horizontal projectionmultiplied by vertical projection reaches the minimum when licenses plate isplaced horizontally or vertically be concerned with ratio of the length and high.The best angle can be fixed by calculate the minimum product and precise locationcan be realized.3, The preprocessing of pictureA series of management to the vehicle licenses plate laid horizontally will befinished in the module about preprocessing of picture including removing frame,separating characters, standardization and getting trait. The main purpose ofremoving frame is to wipe off the influence caused by the nails and the up anddown frames after the image is binaries. The up and down frame is a big obstacle tosegmentation of character string on the licenses plate. Its existence will make thevalue of shadow about characters string are bigger than zero and can not confirmthe point of segmentation by projection so we have to remove it first. The mainidea is the number of the colors changes. There are seven characters on standardlicense plate. The number of the colors changes that is the number of changingfrom black to white or changing from white to black recorded in each horizontalline is less than 14 in the area of character, but almost can not reach 14 in the areahaving no character. The area of character can be determined according to thisfeature .The method is the statistic the number of the number of the colors bouncedof each horizontal line., this line is believed in the area of characters If the numberis bigger than 14 and not in the area of characters if the number is smaller than 14.The topside of these lines is the top edge of characters and the lower line is thelower edge of character if all the lines that may be believed in the area of characterare continuous. Therefore we can confirm the upside downside of characteraccurately and complete the task of removing the frame.The method of vertical project can be adapted to after completing the task ofremoving frame. In order to make the subsequence management convenient andmake the quantity of operation of recognition about program less, each of characterpicture have to be changed into the picture of standardization of position, size andso on. This recognition system standardizes the picture containing 10×20 pixels.The step of getting feature is to pick-up eigenvector which can incarnate thefeature of characters .We adopt the method of scanning pixel in this system4,neural network recognitionNeural network can be train in the way that we put the eigenvector got fromtraining pattern into BP network and the character can be recognized in the waythat put the eigenvector got from the pattern need to recognize into the BP networkhave trained already .BP network adopt structure of three layer .The method ofextracting character adopt the method of scanning pixel. There are 200 points in thelayer for inputting, because the width of standardization is 10 and the high is 20,and then there are 200 characters for each pattern .The middle latent layer will use20 points, and the number of points in output layer is 7.The result of this experiment makes clear that the collection of placeinformation and the recognition of characters on licenses plate can be realizedaccurately. This system has much practical value to the domain of ITS and imageinformatics with simple algorithm and high-speed recognition.