License plate information in license plate recognition and the idea of how to make license plate recognition

First, the current domestic civil license plate are 92-type license plate.

Some parameters of its license plate are as follows:

A: The total length is 440mm;

B: height is 140mm;

C: One of the word Fu is 45mm;

D: The entire license plate area character length is 409mm (here at the back of the calculation when the convenience is considered 410mm);

E: The height of the character is 90mm;

F: The distance between the second and third characters is 34mm;

G: The distance between each of the other two characters is 12mm;

H: If the character "1" is present, the width of the character "1" is 13.5mm, the difference from the other characters is 22.5mm; the distance between two "1" is 38.5mm;

For a 440-long license plate actual in the position of each character in the image;

Second, the whole process of license plate recognition

Individual understanding of the entire license plate processing process should be divided into four pieces;

A: The location of the license plate area, whether you are large background complex image or very clear license plate image, should first locate the license plate area. This should be different depending on how each image is used. (Color space HSI space, morphology of the rectangle discriminant)

B: the license plate of the affine changes, for the previous location of the license plate, see if the character of the license plate is deflected, if there is a need to find the entire character deflection angle, and then to get affine change formula. Then the transformed image is obtained; (Hough change angle, affine change)

C: The character of the license plate segmentation, is a relatively rare piece, if the image is not very good to get the segmentation of this step is more difficult to do. By counting the pixels of each row or column, find the point of separation to determine the position of each character and split it (projection method for character position)

D: The character matching of the license plate, how to recognize the character, there is a template matching method or multi-feature lifting tree (adaboost) or OCR character recognition method to do the neural network is more complex. Template matching is relatively simple, the other two more intelligent;

License plate information in license plate recognition and the idea of how to make license plate recognition