%0 Conference Proceedings
%T License plate character segmentation using Partial Least Squares
%D 2015
%A Gonçalves, Gabriel Resende,
%A Gomes, David Menotti,
%A Schwartz, William Robson,
%@affiliation Universidade Federal de Minas Gerais
%@affiliation Universidade Federal de Ouro Preto
%@affiliation Universidade Federal de Minas Gerais
%E Rios, Ricardo Araujo,
%E Paiva, Afonso,
%B Conference on Graphics, Patterns and Images, 28 (SIBGRAPI)
%C Salvador
%8 Aug. 26-29, 2015
%I Sociedade Brasileira de Computação
%J Porto Alegre
%S Proceedings
%K Automatic license plate recognition, character segmentation, partial least squares.
%X A very important research topic nowadays is the Automatic License Plate Recognition (ALPC). This task consists in locating and identifying an on-track vehicle automatically. This task can be divided into the following subtasks: vehicle detection, license plate detection, characters segmentation and character recognition. This work proposes a new technique to perform character segmentation, which is considered solved in the literature, but in practice is a bottleneck for achieving a robust ALPC system. Our approach is a learning-based technique that uses a regression method known as Partial Least Squares to find the best points where the segmentation should be done between the characters. We perform experiments using a dataset composed of 2,000 license plates and three baselines to compare them with the results obtained by the proposed approach. In addition, we evaluate the usage of the PLS with five feature descriptors and our results show that our method is able to achieve a result up to 46.5% of accuracy, evaluated by the Jaccard measure.
%@language en
%3 2015-Sibgrapi-SegPlate.pdf