| Summary: | 碩士 === 國立成功大學 === 工程科學系 === 87 === Number plate recognition system requires a series of complex image processing steps. In this thesis, a number plate recognition system had been developed. The system includes three phases called pre-processing , ROI(Region of Interest) selection, and character segmentation & recognition. In the first phase, the gray value contrast of the input image is adjusted properly. In the second phase, the area probably contains the number plate is selected by using line based scanning method. In order to avoid the false selection causing by decoration, one to three Region of Interest areas are held. And the scanning is made from left to right, bottom to top. To decide the ROI area, the property of interleaving of black and white pattern for a number plate is used. In the third phase, the characters of the number plate are firstly segmented out by using the line scanning method. The scanning is made from top to bottom of the ROI and accumulated the number of black pixels for the scanned line. The accumulated number is used to evaluate if the line belongs the member of a character. After the segmentation, the character image is binarized and normalized with size 20*10 pixels. Finally, back propagation neural network is used to recognize the segmented characters. 187 data samples are randomly separately into 93 samples for training and 94 samples for testing. Then extra 101 images are used to evaluate the system. The successful rate is about 98%, and the average processing time is 0.7sec for each plate’s recognition.
|
|---|
