Vehicle Number Plate Detection and Character Recognition using MATLAB

Before understanding the concept behind image reading and recognition, you should understand “Fourier Transform” and basic matrix operations in Matlab.

1) Let’s consider we have captured below the picture of the vehicle, in which number plate and its font is clear with the surrounding contrast.

Vehicle number plate detection using MATLAB
Vehicle number plate detection using MATLAB

2) To locate the car license plate and identify the characters, we use the Matlab platform to provide some image processing functions to Fourier transform the character template and the image to be processed as the core idea. The basic method is as follows:

  1. Read the image to be processed and convert it to a binary image. After converting, using the threshold of 0.2 near the license plate characters most clearly, the least points (as shown below)
    I = imread('car.jpg');
    I2 = rgb2gray(I);
    I4 = im2bw(I2, 0.2);
  2. Remove the area where the image is too small to be sure of the license plate.
    bw = bwareaopen(I4, 500);
  3. To locate the license plate, inflate the white area, corrode to unrelated small objects, including the license plate characters.
    se = strel('disk',15);
    bw = imclose(bw,se);

  4. At this point the number plate where the white connectivity domain is clearly visible, but outside the black area, there is a larger white connectivity domain and the license plate where the domain surrounded by. It is necessary to fill it.
    bw = imfill(bw,[1 1]);
  5. Now, let’s find the connectivity domain boundary. While retaining this graphic for later to mark it on it.
    [B,L] = bwboundaries(bw,4);
    imshow(label2rgb(L, @jet, [.5 .5 .5]))
    hold on
  6. Finding the one that is most likely to be the number plate in all connected domains. The standard of judgment is that the aspect ratio of the license plate is about 4.5: 1, and its area is related to the circumference:(4.5 × L × L) / (2 × (4.5 + 1) × L) 2 ≈ 1/27 ,Which is characterized by,

    metric = 27 * area / perimeter ^ 2

    as the matching degree of the connected domain, the closer it is to 1, the more likely that the corresponding connected domain is 4.5: 1.

    MATLAB Program

    stats = regionprops(L,'Area','Centroid');
    for k = 1:length(B)
      boundary = B{k};
      delta_sq = diff(boundary).^2;   
      perimeter = sum(sqrt(sum(delta_sq,2)));
      area = stats(k).Area;
      metric = 27*area/perimeter^2;
      metric_string = sprintf('%2.2f',metric);
    if metric >= 0.9 && metric <= 1.1
        centroid = stats(k).Centroid;
        goalboundary = boundary;
        s = min(goalboundary, [], 1);
        e = max(goalboundary, [], 1);
       goal = imcrop(I4,[s(2) s(1) e(2)-s(2) e(1)-s(1)]);
    for k = 1:length(B)
     boundary = B{k};

    The rectangular matching degree is 0.99, which is the most probable area. Below is the license plate area in the binary image determined by it:

  7. The license plate image is highlighted and expanded to a square of 256 × 256 (left as shown below) so that the matrix rotation operation is performed in the following “Fourier transform”.
    goal = ~goal;
    goal(256,256) = 0;
  8. Read a character template from the file (with “P” as an example, the template image is taken directly from the above binary image). Calculate the Fourier descriptor for the image and calculate the descriptor with a predefined decision function. In the transformed image, the level of brightness indicates the degree to which the corresponding region matches the template (as shown in the figure below).
    w = imread('P.bmp');
    w = ~w;
    C= real(ifft2(fft2(goal).*fft2(rot90(w,2),256,256)));
  9. By checking the maximum value of C, the test determines a suitable threshold (where 240 is appropriate), showing the point where the brightness is greater than the threshold, that is, the highest degree of matching to the template (see below).
    thresh = 240;
    imshow(C > thresh);

    In contrast to the left and right graphs, it can be shown that the character “P” is recognized and positioned. The same way that you can identify and locate other characters.

3) This method is generally easier to understand, Matlab function hides the Fourier transform and other complex calculations. Disadvantages: In the positioning of the license plate, the program specifically according to the characteristics of the given image design, there is no universality. In character recognition, only characters that are substantially consistent with a given template are recognized. License plate size, angle, light, integrity, clarity changes, it can not be identified. At the same time for “8” and “B” similar characters, recognition is often confused.


Click here to download MATLAB program and related files.

If you like this article, please share it with your friends and like or facebook page for future updates. Subscribe to our newsletter to get notifications about our updates via email. If you have any queries, feel free to ask in the comments section below. Have a nice day!

Leave a Reply