Plate Segmentation by useragus


									 A New Algorithm for Character Segmentation of License
                        Yungang Zhang                              Changshui Zhang
             Dept. of Automation, Tsinghua University     Dept. of Automation, Tsinghua University
              The Institute of Information Processing      The Institute of Information Processing
                       Beijing 100084, China                        Beijing 100084,China

                      Abstract                                                  2 Algorithm

                                                           The algorithm has three steps: preprocessing, horizon-
                                                           tal segmentation and vertical segmentation.
Character segmentation is an important step in License
Plate Recognition (LPR) system. There are many dif-
ficulties in this step, such as the influence of image       2.1 Preprocessing
noise, plate frame, rivet, the space mark, and so on.      Preprocessing is very important for the good perfor-
This paper presents a new algorithm for character seg-     mance of character segmentation. Our preprocessing
mentation, using Hough transformation and the prior        consists of size normalization, determination of plate
knowledge in horizontal and vertical segmentation re-      kind and object enhancement:
spectively. Furthermore, a new object enhancement
technique is used for image preprocessing. The experi-          2.1.1 Size normalization: The size of the
ment results show a good performance of this new seg-      plate images is an important factor for the accuracy
mentation algorithm.                                       of character segmentation. All the license plate images
                                                           are normalized to 160*40 in pixel. The experiments
                                                           show that this scale is fit for character segmentation.

                                                                 2.1.2 Determination of plate kind: There
                                                           are three kinds of Chinese license plates: black char-
                  1 Introduction                           acters on a yellow background, white characters on a
                                                           blue background and white characters on a black back-
                                                           ground. The gray scale images are of two kinds: black
                                                           characters on a white background and white charac-
There are many useful applications for a LPR (License
                                                           ters on a black background. The ratios of number of
Plate Recognition) system. The LPR algorithm con-
                                                           white pixels to that of black pixels are quite different in
sists of three steps: license plate locating, character
                                                           these two kinds of gray scale images. So the kind of a
segmentation and character recognition. This paper
                                                           plate image can be determined by histogram analysis.
presents a new algorithm for character segmentation.
                                                           Our posterior segmentation algorithm deals with plate
There are many factors that cause the character seg-       images of white characters on a black background. So
mentation task difficult, such as image noise, plate         in preprocessing the kind of the plate image is deter-
frame, rivet, space mark, plate rotation and illumina-     mined and if a plate image is white characters on a
tion variance. Our algorithm uses Hough transforma-        black background its color will be reversed.
tion and the prior knowledge in horizontal and vertical
segmentation respectively and overcomes the difficul-              2.1.3 Object enhancement: The quality of
ties mentioned above.                                      plate images varies much in different capture condi-
                                                           tions. Illumination variance and noise make it difficult
Compared with the method of image binarization [2],        for character segmentation. Then some image enhance-
this algorithm uses the information of intensity and       ment should be adopted to improve the quality of im-
avoids the abruption and conglutination of characters      ages. As we all know, the image enhancement methods
that are the drawbacks of image binarization. And          of histogram equalization and gray level scaling have
because of using Hough transformation and the prior        some side effects. They may have the noise enhanced
knowledge, the segmentation is more accurate and ro-       as well. For character segmentation, only the character
bust than the simple projection method[1][5].              pixels need to be enhanced and the background pixels
should be weakened at the same time. In fact, a license       space, we obtain n curves in the parameter space. If
plate image contains about 20% character pixels[3]. So        these curves cross the same point (θ0 , r0 ), then the n
these 20% character pixels need to be enhanced and            points in the image space are on a line. So we can find
the rest pixels need to be weakened. It is called ob-         lines in the image space by searching the cross points
ject enhancement. The object enhancement algorithm            in the parameter space.
consists of two steps. Firstly, gray level of all pixels is
scaled into the range of 0 to 100 and compared with                2.2.2 Horizontal        segmentation        using
the original range 0 to 255, the character pixels and         Hough transformation: For plate images with
the background pixels are both weakened. Secondly,            large rotation, it is difficult to obtain horizontal
sorting all pixels by gray level in descending order and      segment lines by horizontal projection analysis. How-
multiply the gray level of the top 20% pixels by 2.55.        ever, for a single character, rotation has little effect
Then most characters pixels are enhanced while back-          on its horizontal projection. It is easier to analyze
ground pixels keep weakened. Fig. 1 shows the result          the horizontal projection of a single character and
of object enhancement. It can be seen from Fig. 1             find the horizontal segment lines. So the horizontal
that after object enhancement the contrast of peaks           segmentation algorithm is as follows:
and valleys of the projection is more significant than
the original.                                                 1) Find valleys of the vertical projection and then ver-
                                                              tically divide the plate image into many blocks. The
                                                              division will not be very accurate because of the influ-
                                                              ence of frame and rivet.

                                                              2) Find the horizontal segmentation line for each block
                                                              by analyzing the horizontal projection of the block. We
                                                              call the horizontal segmentation line for a single block
                                                              a subsection line.

                                                              3) Use Hough transformation on the midpoints of all
                                                              subsection lines to eliminate the incorrect subsection
                                                              lines and combine the correct subsection lines into a
                                                              whole line.

                                                              This method has a number of advantages. First, Hough
                                                              transformation utilizing a vote strategy [6] and the in-
                                                              correct subsection lines are the minority, so the incor-
                                                              rect subsection lines can be eliminated. For example,
                                                              the horizontal segment lines of the block with rivet are
                                                              often incorrect and can be eliminated by Hough trans-
                                                              formation. On the contrary, the linear fitting method
                                                              is more sensitive to the incorrect subsection lines. Sec-
                                                              ond, it is a local projection method, which can weaken
                                                              the influence of background, illumination variance and
                                                              the rotation of plate. Third, it avoids the rotation cor-
            Figure 1: Object enhancement                      rection of images. In fact, rotation correction can cause
                                                              distortion of image and make the character recognition
                                                              more difficult.

2.2 Horizontal segmentation                                   Fig. 2 shows some results of horizontal segmentation.
      2.2.1 Hough transformation: The Hough                   The white lines denote the horizontal segmentation po-
transformation can be used to detect lines in an              sitions. There are images with rotation, background
image[6]. For each pixel in image space (x0 , y0 ), us-       noise, illumination variance, rivet and plate frame influ-
ing transformation,                                           ence in the figure. The results show that the horizontal
                                                              segmentation algorithm has a good performance.

                 r = x · cos θ + y · sin θ

We get a curve r = x0 · cos θ + y0 · sin θ in the parameter   2.3 Vertical segmentation
space (θ, r). Suppose that there are n points in the          The vertical segmentation algorithm is based on projec-
image space. After translating them to the parameter          tion analysis, constrained by the prior knowledge. As
                  (a) Images with rotaiotn

                  (b) Images with noise

                                                                       Figure 3: Vertical segmentation

                                                            3) Estimate the position of the left and right borders of
                                                            the big interval, using the prior knowledge of character
           (c) Images with illumination variance            size. The variance of the pixels’ gray level along a seg-
                                                            mentation line should be small, because a segmentation
                                                            line should be located in the interval of the plate and
                                                            the pixels it crosses are background pixels with similar
                                                            gray level. On the contrary, when it crosses a character,
                                                            the gray level variance will be much larger. Based on
                                                            this fact, the vertical segmentation lines (the left and
                                                            right borders) for the big interval can be retrieved by
                                                            searching around the estimated positions and finding
                                                            the best segmentation lines with the minimum variance
                                                            from the candidates.
            (d) Images with rivet and plate frame
                                                            4) The other vertical segmentation lines can be located
                                                            in the same way.
         Figure 2: Horizontal segmentation
                                                            Fig. 3 shows some results of vertical segmentation. It
                                                            can be seen that the vertical segmentation algorithm
we know, the size of license plate is 440*140(mm), each     can exclude the influence of the space mark and the
character is 45*90(mm), and the interval between char-      plate frame satisfactorily.
acters is 12(mm). And there is a big interval (34mm)
between the first two characters and the last five char-
acters. This information is used as prior knowledge.
And by using the prior knowledge, the segmentation
becomes more accurate. The vertical segmentation al-                       3 Experiment results
gorithm consists of four steps, as follows:

1) Find candidates for vertical segmentation lines. We
assigned a candidate for each valley of the vertical pro-   A database containing 697 license plate images is used
jection.                                                    to test the algorithm. Experiments show that the al-
                                                            gorithm has good performance on character segmenta-
2) Estimate the size of the plate and each character        tion, and can deal with images with disturb of noise,
by using the position information of the horizontal seg-    plate frame, rivet, space mark, rotation and illumina-
mentation lines and the candidates.                         tion variance. Fig. 4 shows some results.
                                                                      4 Conclusion and future work

                                                             The algorithm presented in this paper can segment the
                                                             characters in license plate images accurately. The pre-
                                                             processing can improve the accuracy of the segmenta-
                                                             tion. The algorithm for horizontal segmentation, using
                                                             Hough Transformation, can solve the problem of rivet,
                                                             rotation, and illumination variance. The prior knowl-
                                                             edge constrained vertical segmentation algorithm can
                                                             restrain the influence of plate frame and space mark.

                                                             There are still some further researches to do. For ex-
                                                             ample, it can’t work with some other kind of license
                                                             plate, such as two-rows plate. This problem will be
                    (a) Normal images                        solved in the further work.

                                                             [1] J. Barroso, A. Rafael, E. L. Dagless and J. Bulas-
                                                             Cruz. Number plate reading using computer vision.
                                                             IEEE - International Symposium on Industrial Elec-
                                                             tronics ISIE’97, Universidade do Minho, Guimar˜es,a
                                                             Portugal, Julho, 1997.
                                                             [2] CAO Diming, XUAN Guorong. Abruption and
                 (b) Images with noise
                                                             mergence in character area segmentation. Computer
                                                             Engineering, Oct., 2000.
                                                             [3] J.A. Hegt , R.J. Haye , N.A De la & Khan. A high
                                                             performance license plate recognition system. Proc.
                                                             IEEE International Conference on Systems, Man and
                                                             Cybernetics, pp. 4357-4362, 1998.
                                                             [4] R.A. Lotufo, A.D. Morgan and A.S. Johnson.
          (c) Images with illumination variance              Automatic Number Plate Recognition. IEE Colloquium
                                                             on Image Analysis for Transport Applications, London,
                                                             February 1990.
                                                             [5] Niu Xin, Shen Lansun. Research on license plate
                                                             recognition technology. Measurement & Control Tech-
                                                             nology, Dec., 1999.
                                                             [6] Yuan Jie, Hu Zhengyi, Wang Yanping. Finding
                                                             corners by using Hough transform. Wuhan University
                                                             Journal of Natural Science, Jan., 1998.

                (d) Images with rotation

(e) Plate images of black characters on a white background

        Figure 4: Results of experiments

To top