A Feedback based method for License plate image binarization

Abstract

License plate image binarization is a critical step in Automatic Number Plate Recognition(ANPR) systems and is essential for character segmentation. Generally Otsu(global) or adaptive(local) thresholding methods are commonly used, but each of them may have a shortcoming in terms of segmenting all the characters accurately or Optical Character Recognition(OCR) reading when the plate is not cropped exactly. In this paper we propose a feedback based approach which fuses global and local thresholding methods. Local thresholding method is applied first to extract the character candidates and plate is cropped exactly using the spatial information of the extracted characters. Global thresholding method is applied over the re-cropped plate. Experimental results show the recovery of the number plate reading using the proposed method which are missed or inaccurately read when the global or local thresholding methods are used individually.