Research On License Plates Detection In Complex Background

License plate recognition(LPR) is the important part of intelligent transportation system and is paid more and more attention with development of society and economy. LPR is composed of License plate detection, character segmentation and character recognition, in which, license plate detection is not only the core technique but also the preorder process of the whole system. Meanwhile it affects the performance of overall system. This paper deeply studies the current license plate detection algorithm and does some new trial based on those. The main works in this paper are as follows:(1) This dissertation firstly analyzes image preprocessing technique that corresponds to detecting algorithm. On the basis of analyzing bayonet license plates image, this paper proposed an algorithm of extracting license plates based on multi-information cutting. The algorithm firstly takes full advantage of components’spatial characteristics in the binary image and license plates texture to cut image and shrink searching zone. then, the method of the largest line segment based on projecting is primarily used to locate license plates by adaptive method selection in accurate location. Other than traditional method of searching license plates in the whole of image, this algorithm effectively avoid components that are easily considered as license plate by mistake , on the other hand, it improves algorithm performance.(2) This paper presents a novel method of detecting license plates in complex background based on multi-features. This method is implemented on the basis of blocks, in which, gradient density and license plates feature templates are firstly to express the license plates’feature, then the initial license plates candidates are extracted from image after connective components filtering. After getting initial candidates, two primary relevant filters based on spatial characteristics between license plates and neighbor components are designed to filter out fake initial candidates, in the following accurate location, the method of scanning based on projecting is implemented, finally several effective filtering rules discard fake license plates.(3) A slant angle detection algorithm is implemented based on mathematical statistical data fitting and narrow hole projecting model, the subsequent angle correction is executed through pixels shifting. In addition to this, this paper gradually improve the binarization method in preprocess by experiment and analysis.(4) A mount of experiments and analysis are done for algorithm mentioned above in relevant testing images, the relevant results and performance analysis are presented in corresponding section. The experiment shows license plates detection algorithm metioned above can be effectively used to locate license plates and correct slant angle.