Research On Foggy License Plate Recognition Technology Based On Digital Image Processing

In recent years,with the improvement of the people's living standard,the automobile has become an important means of transportation for the family,and the intelligent traffic management is an important part of the city management.In addition,due to the rapid economic development,fog and haze weather,caused by atmospheric pollution,often occurs.The existence of haze has brought difficulties to the license plate recognition in intelligent traffic management.Therefore,it is significant to study the problem of license plate recognition in fog.Image defogging,license plate location,license plate segmentation and recognition are studied in this thesis.In the field of image defogging,the dark original color prior method and the Yu Jing method are studied.In view of the problem of the partial dark image after two methods for defogging and uneven illumination when collecting images,this thesis combines wavelet decomposition and reconstruction with homomorphic filtering algorithm to solve this problem.Firstly,the dark original color prior algorithm is used to restore the image preliminarily.Secondly,the low frequency component is extracted by using the wavelet transform.Once more,the homomorphic filter is used to enhance the low frequency component.Finally,the image is reconstructed by using wavelet.The experimental results show that the fusion algorithm can enhance the details and brightness of the image.It has a certain accuracy and practical value.In the aspect of image positioning,a series of preprocessing of license plate location is introduced.For license plate preprocessing,the methods of license plate enhancement,edge detection,binaryzation and license plate de-noising are studied.The effects of removing Gauss noise from the wavelet under various threshold values and median filtering are compared.The positioning method of mathematical morphology,projection and region merging is focused on.Firstly,the image is grayscale.Secondly,the gray image is enhanced with special preprocessing.Once more,the rectangular structure is created by mathematical morphology,and the correlation operation is used to eliminate the non-license plate region interference.Finally,the projection method is used to estimate the possible region of the license plate,and the region of the license plate is found according to the length-to-width ratio threshold.The experimental results show that the method obtain good positioning effect.In the aspect of license plate character segmentation and recognition,the methods of license plate correction and character segmentation are studied.The related theories of SVM are elaborated.The methods of feature extraction of license plate and the method of ECOC and SVM license plate recognition are studied.In character feature extraction,the method based on Hog feature extraction is focused on.The extracted features are put into the ECOC and SVM classification models for training and predicting.Finally,the license plate recognition is carried out by using this thesis method in fog and sunny environment,and compared with other methods.The simulation results show that the method used in this thesis has a better recognition rate.