Framework for automatic Malaysian license plate identification system
The requirement of License Plate Identification System differs from one country to another. Malaysian car plates in general appear in different character styles, types (either single or double row), sizes, spacing and character counts and customised printed characters. Such variations cause even det...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2013
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The requirement of License Plate Identification System differs from one country to another. Malaysian car plates in general appear in different character styles, types (either single or double row), sizes, spacing and character counts and customised printed characters. Such variations cause even detecting and localising these plates a difficult problem. The problem of localisation is aggravated further during night time due to poor illumination. To handle these problems, a framework for automatic Malaysian license plate identification system is proposed. In this framework, edge-geometrical features in detecting these plates are introduced. At first, contrast enhancement is applied to increase the detection of edges. The edge part is obtained from the use of Difference of Gaussian operation followed by Sobel vertical edge mask. Then, morphological operations are applied to get the plate region candidates. Using these regions, together with the edge image, geometrical features of these regions are analysed and rule-based classifier is used to correctly identify the true plate region. Next, in character segmentation, normal profile projection and top-bottom feature points are introduced to reduce the problems of connected characters. Finally, extracted characters are then transformed into a single pixel thickness using Zhang-Suen algorithm and recognised using stroke analysis technique. Two experiments have been conducted and compared. The data set contains 250 images captured during day time and 250 images captured during night time. The result of the proposed prototype system shows 92.2% success rate |
|---|