Double-row License Plate Segmentation And Recognition Based On Convolution Neural Network

Automatic license plate recognition technology is used to recognize license plate characters in digital images.The technology mainly includes three important steps: license plate location,license plate segmentation,and license plate recognition.Automatic license plate recognition technology has been widely used in intelligent transportation systems,unattended parking lots and other fields.This technique has practical research significance and application value.Recognition accuracies of existing algorithms on the double-row license plate recognition with Chinese characters are not high.In order to extend the single-row license plate recognition to the double-row license plate recognition with Chinese characters,a double-row license plate segmentation algorithm based on the convolution neural network is proposed,enabling the automatic division of a double-row license plate into two single-row license plates.This algorithm employs the convolution neural network to extract high-dimensional features of license plates and uses them for classification tasks.If the prediction of license plate classification is a single-row license plate,no further processing is required;otherwise,if the prediction of license plate classification is a double-row license plate,it is divided into two single-row license plates with a computed segmentation line.Finally,the character information is recognized by using a convolution cycle neural network structure model.The proposed double-row license plate segmentation algorithm automatically classifies a license plate with Chinese characters,numbers,and letters into two types,and segments the double-row license plate into upper and lower two single-row license plates.The character information from the results of this algorithm can then be recognized effectively by directly using existing recognition algorithms.The experimental results of license plate segmentation and recognition show that the proposed algorithm effectively improves the accuracy of double-row license plate recognition with Chinese characters and is able to produce good results for license plates with poor illumination or blurry characters.Furthermore,the proposed algorithm is based on the training and testing of convolutional neural networks,which can take the advantage of GPU parallel computing and greatly improve the speed of the algorithm.