Assignment Title

CS 585 Project
Ziyang Chen
Xiaoran Xu
11/18/2018


Problem Definition

License plate recognition could be used in a lot of real-world applications like traffic violations, tolls, etc. Most of the license plates has certain pattern, it will probably make the recognition easier. We only need to locate the plate, and do the recognition, but of course there is some processing of the image. A few forseen difficulties are the changing lighting conditions of each image and how we could isolate the plate from the background that might have similar colors, and also the angels of the plate could affect the result of recognition. There are some real-world corner cases to be considered and handled.


Experiments

We will use this dataset, which includes 500 car images, to test our algorithm. And manually confirm the accuracy.

As the recognition of license plate is a relative developed technique in computer vision field, it is not difficult for us to find helpful paper and researches of this topic. After did some background researches, we considered that those four main steps are essential in this project:
1. Preprocessing of the image
In this step, images need to be processed to a more suitable and clearer input for the following steps. Enhancing the constrast, noise reduction and thresholding would be applyed as to bold the characters and sperate numbers from background.
2. Localization of license plates
Vertical prjection and Sobel vertical edge detection would be used to localize the license plates. As some of the license plate in the picture is skew, it would be better to implement Hough transform method.
3. Character segmentation
In this step, Horizontal projection and Sobel horizontal edge detection are appropriate to be used to segment each numbers and characters on the plate. This step is an essential stage for the final recognition.
4. Character recognition
ANN(Artificial neural network) are recommend to be implement in this part. We need to train char data set which contains thousansd of training data from 0 to 9 and A to Z. After that, we could predict the segmented characters on the image.


Expected Result

DNZ 3320
As the image shows above, we want to mark the license plate in the image and output the recognition result on it.


Credits and Bibliography

  • Automated Number plate Recognition System: A. Badr, M.Abdelwahab, A. Thabet, A. Abdelsadek
  • Automated Number Plate Recognition System(ANPR): A Survey: C. Patel, D. Shah, A. Patel
  • Automated Car Number Plate Detection System to Detect Far Number Plates: A. Goyal, R. Bhatia
  • Automatic License Plate Recognition System Based on Color Image Processing: X. Shi, W. Zhao, Y. Shen
  • Survey on Automatic Number Plate Recognition(ANR): K. Sonavane, U. Majhi