Research On Recognition Method For Characters At End Of Steel Billet Based On Character Structure Features

Character recognition is an attractive research in image-processing and pattern recognition field, and its application level and applied value are affected by the recognition precision and speed. How to improve the recognition accuracy and speed of multi-form character of complex scene is a key problem in character recognition research. From the applied project "imaging detection and auto recognition system of characters at end of steel billets", the auto recognition difficulties existed on complex product line for the end of steel billets are analyzed by this paper, such as complex background, various illumination, low contrast, submerged information, blurred characters and short of information. The auto recognition by computer for characters at end of steel billets and how to improve the character recognition accuracy and speed are researched by this paper. The main content is introduced as follows:Firstly, the filter is used to reduce the influence of noise in character recognition at the end of steel billets; for the specialty of steel billet image, the OTSU algorithm is used in this paper to segment the images of steel billet.Secondly, for the existence of multi-line characters at end of steel billets, the accurate localization for multi-line characters is researched by this paper. The character line for recognition is localized accurately by grading localizations based on the edge detection.Thirdly, in order to avoid the influence of inclination characters in localization and segmentation, correction algorithm is researched by this paper, and the tilt character images are corrected by the tilt correction algorithm based on the point operation.Fourthly, with the influence of complex illumination and wicked environment, the quality of the character images at end of steel billets is weak, including the characters with distortion, blurring and deformity. The blur characters with artificial hole are advanced by this paper to extract the features successfully.Fifthly, in order to get the features in favor of character image recognition, the feature extraction and selection are analyzed, some features with good description and discrimination are extracted, and the multi-dimension features are sorted to main features, hypo-features and aided features, which are used to recognize characters. The experiment results testify that the algorithm is effective.