Implementation of k-Nearest Neighbor for PC-Based Character Recognition of Philippine Vehicle Standard License Plate
Authors: Klariz Donna Mae B. Bundal, Arlyn P. Lauron, Jan Jeffrey R. Camiña and Cristina P. Dadula
Abstract
This study focused on the development of a PC- based licensed plate recognition system using Visual Basic programming language. A system that is able to recognize Philippines’ currently used standard vehicle plate numbers using EmguCV image processing and K-Nearest Neighbor machine learning algorithm. The system accepts image as an input or a snapshot of the image from video of a moving vehicle. There were twenty-two (22) unique images of a vehicle in acquired in 3 different positions: upright position (UP), skewed to right position (SR), and skewed to left position (SL). Image processing techniques were applied to the images such as grayscale conversion, Gaussian blurring, and thresholding. Another processing is the detection of plate number area. Optical character recognition is applied to this area where the characters in the image were segmented and individually recognized. The output equivalent characters and the cropped region of the plate number area are displayed on the user interface. The results showed that in: UP, the system recognition accuracy is 83.12%; in SR, it is 39.97%, and 46.21% in SL. The best system accuracy rate was obtained when the captured image of the vehicle is in the upright position which is 83.12%. For better performance, future works may consider the use of exact font style, different angle and position of the license plates, and different lighting conditions of sample license plates for training