License Plate Recognition Based On Deep Learning And Visual Attention

In recent years,with the gradual progress in the research of intelligent algorithms,computer communication equipment,continuous innovation in the field of software,the innovation of science and technology innovation in the field of image processing,so the complexity of the world,is the digital image processing technology is more important.In today's world,people's standard of living is getting higher and higher,along with the rapid development of intelligent products on the market.In this paper,based on the previous research,the license plate recognition system in all aspects of the study,as follows:(1)image preprocessing: analysis of the traditional research methods,this paper combined with the characteristics of the global image enhancement algorithm,because of the noise of the problem has been enhanced,a block image enhancement algorithm.The algorithm of local enhancement of the license plate area is similar to that of the background noise and the introduction of the smaller compared with other methods,especially in the dim exterior conditions,with low contrast images of the vehicles,the effect is very ideal.(2)the location of the license plate: because in the complex background color image in the license plate location of low accuracy,according to the color characteristics of the license plate and the specific characteristics of China license plate,this visual attention model of vehicle license plate location based on.The competition mechanism of the license plate color is the color feature of the image(red green and blue yellow feature map feature map),combined with various directional filtering feature map,integrated object detection multiple low-level features,thus forming a saliency map.Combined with the information of the license plate,select the license plate image.The method has the advantages of high accuracy and good fault tolerance.(3)the license plate image of the two values: the license plate characters have complex light characteristics,through the license plate image illumination compensation,and the use of the two value algorithm.Firstly,the influence of uneven illumination on the license plate is solved,and then the two value algorithm is used to compensate the background of the license plate image.After being compensated by illumination,the influence of illumination on the license plate is greatly reduced,which indicates that the two value is higher.(4)the character segmentation: because the license plate location after the influences of noise,positioning of license plate tilt of the characters will also have some impact,but this process is inevitable,is proposed in this paper.The correction method of license plate region,this method can be a good solution to the distortion character of the fracture so,after correction of character segmentation.(5)the recognition of the character: according to the particularity of the Chinese license plate character,the character recognition is a key step in the license platerecognition system.In this paper,the method of feature extraction and deep learning network is used in the classification technology.The rough grid feature selection and feature recognition as an important character,through neural network training,the training is easy to confuse the character input to the neural network,especially the complex structure according to the characteristics of partial difference Chinese characters,subtle;and on the control of standard samples,this method can improve the recognition efficiency.