Local Enhancement of Car Image for License Plate Detection


In this paper we address the problem of car plate detection. In the first part of algorithm we propose a method for en-hancing car plate regions. We estimate the local density of vertical edges in the image as a criterion for local enhance-ment. In the second part, the vertical edges from the en-hanced image is then extracted and feed to a morphological filter to constitute candidate regions for the place of car plate. This filter aims to connect vertical edges closer than the size of a defined structuring element. The output of this process is a number of connected components among which the plate region is. We use some geometrical features of car plate such as shape and aspect ratio to filter out non-plate regions from the candidate list. Using the correlation be-tween the candidate regions and the model of car plate, the most probable region is found. The result of enhancement on car images shows the ability of the proposed method for im-proving the contrast of plate region(s) in the image. The ex-perimental results on out-door car images confirms the ro-bustness of the proposed method against the severe imaging condition.

Download full-text


Available from: Vahid Abolghasemi
  • Source
    • "There are many techniques have been proposed to address vehicle license plate location. There are some methods of them based on mathematical morphology and edge detection analysis [1-15], such as based on histogramming and mathematical morphology [1], vertical edge detection and mathematical morphology techniques [2] [4] [6], edge detection and mathematical morphology for yellow license plate [3], hat transform and morphological operations [5], dynamic RGB threshold formula by limiting the hue change range and mathematical morphology [7], morphology and auto–correlation [8], horizontal gradient and morphological operation [9], edge finding and window method [10], line detection method to detect straight lines and weight assignment scheme to obtain a weight and regions with the densest edges to select candidates [11], topology of characters and outer contour [12], edge analysis and statistics [13], multistage approach for analysis of vertical edge gradients from contrast stretched gray-scale images [14], edge detection, morphology operation and color analysis [15]. But most of previous methods have some restricted in cases: uncertainty of edges, various types of plate, broken edges, the plate is small, dim lighting, the images are fuzzy, distorted or hided and fails in situations such as low or high illuminated images. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The License Plate Location (LPL) from the vehicle images is the key step in the Automatic License Plate Recognition (ALPR). In this paper we proposed a new LPL algorithm for Vietnam license plates, which combined preprocessing, morphology operation on grayscale image, image subtract operation on grayscale image, image binarization based on threshold, edge detection use Canny operator, morphology operation on binary image, finding the license plate (LP) angle & rotating LP based Radon transform and bilinear interpolation, and then cutting exactly license plate region based on measuring properties of Vietnam license plate regions. Specially, by the combined morphology operation & subtract operation on the grayscale image, which very efficiency for complex background images, night & day images and different illumination, different light conditions, we have obtained the better image with new intensity values, which more satisfied for the image binarization and reduced candidate regions. And we used opening operation on binary image to remove small objects (noise areas), and by the measuring properties of license plate regions we had cut exactly for different Vietnam license plate dimensions (one-row and two- row LP). So, the proposed approach is more effective than some of the existing method earlier and very satisfied with Vietnam license plates. Experiments have been implemented on 500 images taken from different background: lightening conditions (night and day), license angles, illumination, size and type of license plates, colors, reflected light, dynamic conditions. The efficiency of processing of the proposed algorithm is improved and average rate of accuracy of the LPL is 97.27%, and proposed method is suitable for all of color license plates.
    Preview · Article · Jul 2011
  • Source

    Full-text · Conference Paper · Feb 2014