الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 126 |  |  |
| | | from | 126 | | |
Abstract
In this thesis, a system for license plate recognition is proposed. This system consists of three steps: License Plate Detection (LPD), character Segmentation (CS), and character Recognition (CR).In LPD step, the information in images is filtered by locating vertical edges using Sobel vertical mask and then the image contrast is enhanced using contrast stretching. Then, The cluttered edges are reduced by using the horizontal projection where any row has white pixels less than 35% of the maximum value in the horizontal projections is omitted. After that, we try to get all connected components using a series of mathematical morphology operations such that, we try to get the coordinates of the smallest rectangle that contains only the plate as possible as we can. Finally, we determine the LP from the candidate regions by classifying them using trained support vector machine not by height, width, aspect ratio, and color, to make the proposed system more efficient and reliable especially the used images have many regions which have the similar aspect ratio, color and also contains a text.In the CS step, the tilt is corrected, salt and pepper noise are removed. Then an adaptive threshold algorithm is used to binarize the LP image. After that, all small objects are removed following by extracting all connected components using Connected Components Labeling (CCL) technique. Finally, all non-characters components are removed using the aspect ratio and the position of these objects.Finally, in the last step of the proposed system, the extracted characters are classified using two different techniques Template matching and support vector machine with the histogram of orientation gradient descriptor. Then, the recognition results using both techniques are compared and discussed. The proposed system is implemented using Matlab and tested on a dataset contains images for buses, vans and cars captured in varying weather conditions, different day times, different capturing distances and angles, and contains many regions that are similar to the LP and have a complex background. The simulation results reveal the efficiency of the proposed system.