Extraction System of Characters in Vehicular Plates

In this paper, is proposed a system that identifies the location and dimensions of the characters relative to the image of a vehicular plate, when the location has not been accurate. The system is divided into 4 stages, each with a specific purpose. Which are: binarization by thresholding, morphological filtering, identification of the largest area and segmentation similarity. The first stage is used to find at the extent of possible, the region occupied by the plate relative to the rest of the image. Then, in the filtering step, it seeks to eliminate as far as possible the noise that interferes with the identification of the characters. The third stage is used to identify the region occupied by the plate. Finally, segmentation by similarity is used to identify position and dimension of the characters at the image. This stage is implemented by a Kohonen neural network.