Morphology-based Train Compartment Number Extraction

Abstract

A morphology-based train compartment number extraction algorithm was introduced to provide an accurate and effect service to subsequent recognition process. First, a preprocess which included gray-scale transformation, image normalization, morphological closing operation filter, OTSU threshold segmentation was applied to the input image. Second, an extraction process to train number followed, which included a coarse character selection scheme, a union-find-based characters cluster method and a fine characters selection procedure, to form train number region candidates. To the end, a train number recovery algorithm was proposed to recover the boundary of the candidate with left-expansion and right-expansion so that all missed train number characters could be recovered. In this paper, the proposed was applied to 100 train carriage images for testing. The experiment results show that the proposed algorithm is robust and is adaptable to the illumination conditions, train number position, tilt, background and font, and 95% recognition ratio is achieved, which improves much the validity of recognition process.