Research On License Plate System Based On Improved Dark Primary Color Channel Under The Fog-Haze Environment

In recent years,frequent large-scale foggy weather in China has seriously affected people’s daily life and work.Foggy weather will lead to the reduction of contrast,saturation and color degradation in the target image,which will lead to the degradation of the recognition effect of the image recognition system,such as the monitoring of intersection traffic,residential license plate recognition,highway license plate recognition,and even lead to the system paralysis.Therefore,image sharpening in foggy weather is of great practical significance.Vehicle image collection due to fog,is unable to locate the location of license plate.Foggy weather is sweeping large and medium-sized cities in China,resulting in color distortion,low contrast and blurred images when collecting vehicle license plates in foggy days.It directly affected the real-time and precision of license plate recognition system.It can be seen that pre-processing vehicle image in foggy weather is an important problem for license plate recognition in foggy environment.In this thesis,histogram equalization algorithm,retinex algorithm and dark primary color defogging algorithm are respectively implemented by theory and python+opencv environment.Laplace gradient algorithm is adopted as the criteria for clarity evaluation.Finally,the dark primary color defogging algorithm is the best.In addition,the median filtering algorithm is used to optimize the dark primary color defogging algorithm with high time complexity and white edges.The median filter is used to improve the HE dark primary color defogging algorithm.The median filtering improves the white edge effect on the original basis,avoids the loss of a large amount of time caused by the optimized transmittance diagram in the dark primary color defogging phase,and adjusts the super saturation problem in the image.Canny algorithm is firstly used to detect the edge of foggy images,and morphology is used to detect the edge data of foggy images,so as to obtain the corresponding closed space and obtain the candidate position of the corresponding license plate according to the fixed length and width ratio of the license plate on the vehicle.After that,use vertical projection to process license plate candidate area,according to this area display clear valley and peak alternating run this function,accurate positioning license plate area.Finally,color features are used to eliminate interference factors.Character segmentation uses projection.Character recognition uses mature SVM algorithm for recognition.The program implementation environment is python + opencv.The main work of this thesis is to analyze and compare the main defogging algorithms to get the best defogging algorithm,and propose improvements to the deficiencies of the current algorithms.The effect of defogging is achieved by preprocessing the acquired images,and the license plate recognition system is implemented.