Research On License Plate Recognition Based On Retinex And Non-undersampled Contourlet Image Enhancement

With the rapid development of the economy,the continuous improvement of the quality of life,the frequency of using cars is getting higher and higher,and the license plate recognition system is also making rapid progress.In real life,the captured image is not ideal,so it is very difficult to identify the image when it is implemented,and it is easy to identify mistakes.Compared with other image recognition systems,the requirements for timeliness and accuracy will be more stringent.Although domestic and foreign scholars have done a lot of research,there are still defects in the license plate recognition technology in complex environments.This paper firstly studies the current situation of license plate recognition technology at home and abroad,and proposes the improvement of license plate recognition algorithm based on the existing algorithms.This paper mainly aims at the improvement of the preprocessing and character recognition of the license plate system,and implements the license plate recognition system based on the improved algorithm.The main work and innovations of this paper are as follows:Firstly,the image enhancement of the license plate preprocessing stage is proposed.For the problem of insufficient contrast and overall darkness in some license plate images under severe weather conditions such as haze,an improved image enhancement algorithm based on the non-downsampling domain is proposed.The improved license plate image preprocessing algorithm can significantly improve image contrast and information entropy,thus improving the visual perception of the image,and effectively enhancing the image in the preprocessing stage,providing early conditions for subsequent character segmentation and recognition processing.The second is to improve the license plate character recognition.After the preprocessing and character segmentation are completed,the current license plate character recognition algorithm is analyzed to find the shortcomings of network training failure due to the large sample space classification,and the convolutional neural network.The gradient back modification method can directly overcome the above deficiencies,so the license plate characters are recognized.According to the algorithm theory,a six-layer convolutional neural network structure is proposed,which can overcome the shortcomings of training failure and large delay.Experiments on the algorithm show that the improved algorithm can significantly improve the recognition rate.Thirdly,combined with the improved algorithm of the above preprocessing stage,the license plate recognition system is implemented on the platform,which can effectively process the captured dark and noisy license plate image and obtain the correct recognition result.Figure[30]Table[4]Reference[55]...