0

I am developing the Russian system ANPR (full open sourse) .

The system already knows how to find the license plate.

But I am having difficulty with the division of numbers and letters in the license plate.

I know the method Haar. Do I need to use it to achieve this goal or are there other methods?

Image number 1 shows that I have. Image number 2 shows that I need.

enter image description here

KostyaKulakov
  • 25
  • 1
  • 5
  • 2
    [read a bit](https://github.com/MasteringOpenCV/code/tree/master/Chapter5_NumberPlateRecognition) for some more ideas – berak Jun 27 '14 at 15:49
  • 1
    See https://github.com/openalpr/openalpr might be helpful... – Haris Jun 27 '14 at 16:12
  • Haris, thanks for answer. i looked this is project. – KostyaKulakov Jun 27 '14 at 17:06
  • 1
    It might be useful: http://stackoverflow.com/questions/10964226/how-to-convert-an-image-into-character-segments – Abid Rahman K Jun 28 '14 at 11:21
  • harr method will be of little use for segmenting characters. as per @AbidRahmanK, adaptivethreshold as first step is a very effective way to get characters. after that, you can define some country specific optimizations. try to describe the relationship between characters. in this case, first character is always two third the 2,3,4th characters for example. then out of all possible contours, choose the combination that have the lowest deviation. you can also try horizontal and vertical projection for segmenting chars after deskewing the plate. – Zaw Lin Jun 29 '14 at 15:27
  • another property you can use is the distance between centers of each consecutive characters. if you calculate the standard deviation, it should be very low. you can use this to filter out most of the outliers/noise or as a ranking score when you are evaluating the contour combination as i have described above – Zaw Lin Jun 29 '14 at 15:32

0 Answers0