Research On Layout Understanding And Fast Recognition Of Ballot Image

Democratic election is the most powerful democratic instruments in the history of mankind; it is an important process for people to exercise their rights. Ballots which are the instruments on which the democratic election depends for existence are the intermedium for people to express their hopes and exercise their rights. But it's an urgent problem to realize the Ballot Count of various kinds of ballots with huge amount efficiently.Based on analyzing the research status of the most reasonable method in home and abroad—Ballot Count based on OCR, we can find that this method has many defects in low automation, low recognition speed, low recognition accuracy and the narrow scope of application of ballots types. Aimed at these shortcomings mentioned above, the main tasks we have done are as follows:1. A method based on run-length of table line to realize the geometry structure recognition of ballot image is present. The geometry structure recognition of ballot image can come down to the geometry structure recognition of ballot table. So we can realize the geometry structure recognition of ballot image by detecting and analyzing table lines in the ballot image. The run-length of table line has the advantages of easily detecting, restoring and processing, and it perform robustly to detect table lines by run-length of table line. The situations of wrong recognition and missed recognition of table lines are reduced, and the recognition speed is also improved.2. A method based on a binomial tree model to realize the layout understanding of ballot image is present. The layout understanding of ballot image can come down to the geometry structure recognition of ballot table and the logical structure recognition of ballot table. We have defined many logical structures of ballot table, such as the candidate information field, the voting symbol field, the voting unit, and the voting row (column). With these logic structures of ballot table, we construct the logical model of ballot table. And finally we use a binomial tree model to realize the logical structure recognition of ballot table. The method realizes the machine self-learning of layout of regular ballot image efficiently.3. A reasonable recognition scheme of ballot image is present. And this recognition scheme not only improves the recognition accuracy and the recognition speed of ballot image, but also is suitable for many kinds of ballots types.4. The "Intelligent Ballot Count System Based on OCR" development practice was illustrated, which confirms the effectiveness of the methods mentioned above.