Paper Info
Title
Classifying License Plate Numerals Using CNN.
Abstract
Nowadays, security cameras are usually set in various places in Japan. The cameras are effective for criminal investigation. Especially, a license plate on a car which is in images by the cameras can identify the car. However, numbers on the license plate photographed by the cameras sometimes unreadable for humans since the image of the numbers is often poor picture quality, and noise and light decrease the quality much more. Therefore, we propose a new method to read numbers on a license plate with poor picture quality and we evaluated our method by experiments. In this paper, we described the method, experiments, evaluation and plans in future. The main idea is to read the numbers by machine learning on CNN which a lot of images of numbers created by three dimensional rotations and retouching are put in. The retouching processes in this paper are shift, cropping, smoothing, noise assignment, brightness changing and random erasing. A model created by the learning with the created images is saved and used for the classification of numbers on license plates. We think that the method is technically new since we have never heard the method to use three dimensional virtual numbers for the classification of numbers on real license plates. We prepared photos of real license plates and experimentally classified them by decreasing their resolution in stages. As a result, images with only 2 by 4 square pixels resolution were able to be classified with a probability of 99%. On the other hand, the same image with different cropping area was sometimes classified with a quite low probability. We will identify the cause of the problem in future.
Year
DOI
Venue
2019
10.1007/978-3-030-19063-7_84
PROCEEDINGS OF THE 13TH INTERNATIONAL CONFERENCE ON UBIQUITOUS INFORMATION MANAGEMENT AND COMMUNICATION (IMCOM) 2019
Keywords
DocType
Volume
License plate,Classifying,CNN,Machine learning,Deep learning
Conference
935
ISSN
Citations 
PageRank 
2194-5357
0
0.34
References 
Authors
0
2
Name
Order
Citations
PageRank
Tomoya Suzuki1243.37
Ryuya Uda26321.76