The Research Of Business Card Image Segmentation And Recognition

This paper focuses on English business card image segmentation and recognition project and presents corresponding algorithms. We first introduce the main framework of English business card image segmentation and recognition, including image preprocessing, document image analysis, feature extraction, classification and character recognition processes, then introduce and describe every function module and process of main framework and compare related algorithms.At stage of character segment process, this paper firstly segment character images on connected component technology. Then, the paper proposes a water reservoir scheme which is mainly based on character features to segment connected character image. Water reservoir topological structure and attribute denotes key information of space between two characters which may provides key basis of character image segmentation. The related tests approve this method can segement connected character image correctly in certain degree.At stage of character recognition process, this paper combines pattern match algorithm and BP neural network method. At first, the paper will use pattern match algorithm based on improved weighted Hausdorff distance. Then, character feature extraction for BP algorithms input vector will be carried out, including structural feature and characteristic feature extraction, each has its merits and faults. So we combine both of them to construct character features. The experiment shows that the proposed method validates improvement of character recognition.