License plate detection using point of interest detectors and descriptors

A license plate detection task (LPD) is an important stage in the License plate recognition (LPR) implementations for automated transport systems. This paper is focused on a theory behind various interest points (IPs) detectors and region descriptors often used in applications of image processing including the LPR apps among others. We have carried out wide analysis on behaviour of several selected detectors and descriptors listed here: Harris, Shi&Tomasi, MSER, FAST, SURF and BRISK. These algorithms have been implemented, tested and compared each other from a point of view of their stability and repeatability. Our experiments were performed on a license plate localization task and carried out over an extensive gallery of real vehicle images.