Research And Implementation Of Fuzzy License Plate Recognition Algorithm

With the continuous development of the social economy,the number of motor vehicles is also increasing.As a result,the intelligent transportation system has emerged,and its core is inseparable from the license plate recognition technology.However,in intelligent surveillance video,vehicles tend to move quickly,so that the captured images are often ambiguous,which adds difficulty to later work such as analysis and recognition of targets.In this paper,the algorithm of license plate recognition is studied.The focus is on the restoration of license plate image and the recognition of character after restoration.The main research contents are as follows:（1）Aiming at the problem of deblurring for license plate image,a restoration method based on random sampling consistency（RANSAC）is proposed.Firstly,the Canny edge detection is performed on the blurred image in cepstrum domain.And the RANSAC method is used to estimate the blur direction.Then,the blur length is estimated by the zero component of the blurred image in cepstrum domain.Finally,combining the advantages of Richardson-Lucy（RL）algorithm and L₀regularization priori,the blurred license plate image is restored.（2）In the aspect of license plate location,the restored image is pre-processed with gray enhancement and median filtering.Then the license plate itself and the characters rich edge information are used to perform Sobel edge detection on the pre-processed restored image.Morphological operations such as expansion,measuring and extracting the corresponding shapes in the image through structural elements.Finally,the size and shape characteristics of the license plate will be used to select the contour area to determine the license plate area.（3）In the aspects of license plate correction and character segmentation,the reasons for the license plate tilt and the type of tilt are analyzed.Several common tilt correction methods are introduced,and the correction method based on Radon transform and rotary projection method is adopted.At the same time,some common character segmentation methods are introduced.The advantages and disadvantages of these methods are analyzed,and finally the segmentation based on projection and character features is used.（4）In the aspects of character recognition,in order to improve the recognition rate of license plates,this paper proposes a recognition method based on combined features and BP neural network.Firstly,considering the confusing characteristics of numbers and letters,LBP is used to extract local texture features.And then according to the characteristics of license plate Chinese characters,LBP operator is improved,improved LBP and horizontal vertical projection are used to extract features.Finally BP network is trained with extracted features and used to identify license plate characters.