| dc.description.abstract | Business cards usually convey two kinds of information. One is the personal information, such as holder’s name, address, telephone number, e-mail address, etc. The other is the information of the company, such as company name, logo, address, telephone number, etc. In business cards, the relationship between company name and logo is one-to-one mapping. Thus, if we can recognize the logo, we can also know the company name of the business card. The goal of this thesis is to acquire the company name of any business card by means of extraction and recognition of its corresponding logo. Once we have acquired the company name of the business card from its corresponding logo, we can use it to correct the OCR result of company name in the business card. Besides, the formats of business cards are usually the same for the same companies. Thus, we can use the layouts that are extracted from the business cards to classify them.
Extraction and recognition of logo are the focus of this thesis. In the extraction stage, an input business card is first segmented into many homogeneous blocks and each block is then given some attributes. Some rules are constructed based on the extracted attributes to extract the logo. In the recognition stage, the extracted logo is first normalized into and then transformed into matrix of wavelet coefficients. In our work, we use the coefficients of index [0,0] and 40 largest-magnitude as the features. Finally, the features of the considered logo are compared with those stored in logo database to obtain the recognition result.
In our experiments, 90 business cards are tested. Some are Chinese formats and some are English ones. Experimental results reveal the feasibility and validity of our proposed method. | en_US |