Segmentation of license plates through clustering techniques in external environments

This paper presents the use of clustering techniques to segment license plates through a method applicable to systems ANPR (Automatic Number Plates Recognition). To achieve this goal were different tests with clusters, comparing its performance and effectiveness. This method uses clustering techniques to remove automatically, spurious points that can be generated by the correlation techniques. To perform segmentation of license plates, the method receives as input a cloud of points corresponding, resulting in a matching between the input image descriptors SIFT (Scale Invariant Features Transforms) and the reference image descriptors SIFT that contains the possible characters of a license plates. Through clustering techniques, gets the cluster consisting of the points located in the zone of the plate and also his associate centroid. Regions growing technique is then used to segment the minimum area that encapsulates the centroid which constitutes the patent plate of the considered image. In conclusion, we present a clustering process that is part of a hybrid approach that combines techniques from computational vision and intelligence, to segment license plates in external environments.