Subhradeep Kayal
License Plate Detection, License Plate Extraction, Gabor Filter, Cross-cut
Vehicle license plate location is a key factor of the vehicle plate recognition system. This paper aims at developing a novel method for locating license plates, from both complex and close-up images, based on extracting areas with high vertical edge density and variance parameters. In this method, the input image is first convolved with a 2-dimensional Gabor Filter with a vertical orientation so as to filter out the vertical edges. Areas of the image with a high vertical edge density, and structure and texture similar to that of a license plate region, are extracted. The next step is variance based filtering, motivated by the knowledge that license plate regions are generally areas of ‘change’, and hence have a high value of variance. Finally, a method of ‘cross-cuts’ is adopted to effectively separate the actual plate region from any region with characters or letters which may not have been separated during the variance filtering. The algorithm was tested on 139 images with a rate of location of 97%, proving that it is effective and competitive. It has been found to be robust to environmental variations such as camera positioning and poor lighting conditions.
Important Links:
Go Back
