Download License Plate Recognition System-Computer Sciences Applications-Project Report and more Study Guides, Projects, Research Applications of Computer Sciences in PDF only on Docsity!
ii
Table of Contents
Table of Contents................................................................................................................ ii
List of Figures .................................................................................................................... iii
iii
- 1 Previous Work Abstract iv
- 1.1 LICENSE PLATE EXTRACTION (HOUGH TRANSFORM)
- 1.1.1 Assumptions.......................................................................................................
- 1.1.2 Simple Block Diagram.......................................................................................
- 1.1.3 Gray Scale Conversion
- 1.1.4 Edge Detection...................................................................................................
- 1.1.5 Hough Transform...............................................................................................
- 1.1.6 Candidate Verification
- 1.2 DRAWBACKS
- 1.3 CONCLUSION
- 2.0 Current Work
- 2.1 LICENSE PLATE EXTRACTION (EDGE STATISTICS )
- 2.1.1 Modified Block Diagram
- 2.1.2 Preprocessing
- 2.1.3 Image Enhancement...........................................................................................
- 2.1.4 Connected Component.......................................................................................
- 2.1.5 Filtering Connected Components
- 2.1.6 Candidate Verification
- 3 System Configuration
- 4.0 Results.........................................................................................................................
- 5.0 Comments
- References.........................................................................................................................
- Figure 1 A simple block diagram for Extraction Phase...................................................... List of Figures
- Figure 2 A successful case Figure 3 Unsuccessful case.
- Figure 4 Modified Block Diagram......................................................................................
- Figure 5 Edge Detected Images
- Figure 6 Candidates after applying aspect ratio..................................................................
- Figure 7 Original Colored Images (Successful)................................................................
- Figure 8 Extracted License Plates.....................................................................................
- Figure 9 Original images with half extracted License plates............................................
1 Previous Work
1.1 License Plate Extraction (Hough Transform) In order to identify a vehicle by reading its license plate successfully, it is obviously necessary to locate the plate in the scene image provided by some acquisition system (e.g. in our case still camera). Locating the region of interest in the scene aids in dramatically reducing both the computational expense and algorithm complexity. For example, an image of resolution 640x480 which contains the region of interest (in this case a license plate) may account only 10% of the image area for region of interest.
1.1.1 Assumptions
To extract license plate from the image the assumptions are A visible boundary should be present between license plate and vehicle. License plate is of rectangular shape. Captured image contains only one and complete license plate attached to vehicle. License plate is clearly visible in the captured image.
1.1.2 Simple Block Diagram
The block diagram of the extraction phase of license plate is shown below.
Figure 1 A simple block diagram for Extraction Phase
Hough Transform CandidateRegions
Colored Image Gray scaled Image
Enhanced Image
Edge Detected Image
Extracted License Plate
Candidate Verification
Hough Transform CandidateRegions
Colored Image Gray scaled Image
Enhanced Image
Edge Detected Image
Extracted License Plate
Candidate Verification
Colored Image Gray scaled Image
Enhanced Image
Edge Detected Image
Extracted License Plate
Candidate Verification
1.1.3 Gray Scale Conversion
In the start of this phase we have to first convert a colored image to a grayscale image and than other techniques for edge detection is applied on that. This step is performed to overcome a very general difficulty that all the number plates may not be of the same color and all the characters on the number plate may not be of the same color. RGB/colored images can be converted to grayscale by eliminating the hue and saturation information while retaining the luminance.
1.1.4 Edge Detection
An extremely noticeable characteristic of license plates is their rectangular shape. Edge detection is used to detect this rectangular shape. An edge is the boundary between two regions with relatively distinct gray-level properties. In order to detect the borders or edges in an image it is useful to focus on abrupt changes in intensity in local neighborhoods. Edge detecting an image significantly reduces the amount of data and filters out useless information, while preserving the important structural properties in an image. Difference (derivative) operators produce high values as a response to rapid changes in gray level and are therefore useful as edge detectors. In edge detection process it takes an intensity or a binary image I as its input, and returns binary image BW of the same size as I, with 1's where the function finds edges in I and 0's elsewhere. The edge detector used here is Sobel edge detector, although Canny provide better results than Sobel by giving every small detail but as we are not focusing on small objects in case of plate extraction, so Sobel best suites our application.
1.2 Drawbacks
Edge linking process cannot be generalized for all the images in database. When linking edges to form a complete rectangular shape of a license plate, characters on plate are also linked. Processing time for Hough transform is very large (3-5 min for an image of resolution 640x480 on 800 MHz processor Intel).
1.3 Conclusion
The above discussed technique cannot be used because the results achieved by applying Hough transform cannot be generalized for all the images in database.
Now we focus on a technique called Edge Statistics and Morphology to be used for locating license plates in images. The extraction phase is started again based on this approach which is discussed below.
2.0 Current Work
2.1 License Plate Extraction (Edge Statistics)
The technique now used is Edge Statistics and Morphology, based on the characteristics and shape of edges (Morphology), the assumption in this case are the same as discussed above.
2.1.1 Modified Block Diagram
Figure 4 Modified Block Diagram
Colored Image Gray scaled Image
Edge Detected Image
8-Connected Components
Height and Width Ratio(1:1 to 1:5)
Extracted License Plate
Candidate Regions
Length<ImgLength/
Candidate Verification
Least smooth Region
Image Enhancement
Thickening Algo
Colored Image Gray scaled Image
Edge Detected Image
8-Connected Components
Height and Width Ratio(1:1 to 1:5)
Extracted License Plate
Candidate Regions
Length<ImgLength/
Candidate Verification
Least smooth Region
Image Enhancement
Thickening Algo
2.1.3 Image Enhancement
The edges detected by the Canny edge detector may not be connected in some of the cases, so it will be difficult to completely extract the License Plate if it is not fully edge detected. To over come this difficulty a thickening algorithm has been employed here. It thickens objects by adding pixels to the exterior of objects until doing so would result in previously unconnected objects being 8-connected. The result of applying this algorithm is that small gaps between edges are filled and we have 8- connected components in last. Thickening is done by thinning the background of the set in question and than complement the result. In other words, to thicken a set A, we form C= A c, thin C and then form Cc.
2.1.4 Connected Component
Next we have to analyze the connected components for 8-connectivity. To establish if two pixels are connected, it must be determined if they are neighbors and if their gray level satisfies a specified criterion of similarity (say if their gray levels are equal). Connected component labeling works by scanning an image, pixel-by-pixel (from top to bottom and left to right) in order to identify connected pixel regions, i.e. regions of adjacent pixels which share the same set of intensity values V. (For a binary image V={1} ; however, in a gray level image V will take on a range of values, for example: V={51, 52, 53, ..., 77, 78, 79, 80} .)
The connected components labeling operator scans the image by moving along a row until it comes to a point p (where p denotes the pixel to be labeled at any stage in the scanning process) for which V={1}. When this is true, it examines the four neighbors of p which have already been encountered in the scan ( i.e. the neighbors (i) to the left of p , (ii) above it, and (iii and iv) the two upper diagonal terms). Based on this information, the labeling of p occurs as follows:
If all four neighbors are 0, assign a new label to p , else If only one neighbor has V={1} , assign its label to p , else
If more than one of the neighbors have V= {1} , assign one of the labels to p and make a note of the equivalences.
After completing the scan, the equivalent label pairs are sorted into equivalence classes and a unique label is assigned to each class. As a final step, a second scan is made through the image, during which each label is replaced by the label assigned to its equivalence classes. For display, the labels might be different gray levels or colors. Two pixels p and q are said to be connected in S if there exist a path between them consisting entirely of pixels in S.
In our case the edge detected image contains pixel with value ‘1’ where it find edges and ‘0’ else where, so in this 8-connectivity is considered between pixels with value 1.Inlast we have only 8-connected components which are labeled from 1 to n, where n is the maximum number of connected components (1st^ connected component is labeled as 1, 2nd^ as 2 and so on.)
2.1.5 Filtering Connected Components
The approach here is to extract License Plate using aspect ratio because License Plate is of rectangular shape. In this step the connected components are analyzed for their height and width ratio to fulfill the rectangular shape criterion. The candidates with height and width ratio within 1:1 to 1:5 are considered only. This range of candidates is selected because the database of images does not contain plates of some specific standard and but plates whose aspect ratio is within this range.
Area = Length x Width
So by 1:1 to 1:5 we means the areas whose length width ratio is 1:1 to 1:5 (means length is 5 times width).
After applying this constraint on the whole image we have 4-7 candidates which includes the license plate, indicators, wind screen, and sometimes some other objects which also fulfill this criterion.
Figure 7 Original Colored Images (Successful)
Figure 8 Extracted License Plates
Figure 9 Original images with half extracted License plates
3 System Configuration 800 Mhz Intel Original Processor. 256 MB RAM. Matlab 7..
4.0 Results
The system is tested on database of 100 images which include Pakistani and British icense Plates. License Plate Recognition Rate is 91 %. conds. 6% of plates were not correctly recognized (plates were half extracted). 3% of plates were not found.
s
he license plates which were not successfully recognized were due to the fact that e contrast between the plate border and color of vehicle was not strong enough to enerate an edge, and also some of the plates were half extracted due to incomplete dge detection of the plate. This incomplete edge or broken edges problem has been tried to overcome by using the thick ps by adding of object until it
L
Processing time is 2 se
5.0 Comment
T th g e ening algorithm which fills small ga pixels on the exterior boundary is 8-connected.
