Research On Several Key Technologies Of Ballot Recognition Based On Layout Understanding

With the continuous improvement of the democratization and the perfection of the democratic system in China, there are more and more election activities. Generally, there are four kinds of voting schemes:manual voting, electronic voting, OMR-based voting, OCR-based voting. However, there are still some problems which need solving, such as time-consuming, limited application scope, lack of intelligence and security. A voting approach based on layout understanding is researched. Several key technologies of traditional ballot image recognition are researched in the paper. A rapid vote-counting system based on layout understanding is developed. The main contents and contributions are as follows:(1) The geometry structure recognition of ballot table is researched.Firstly, the layout characteristics of ballots are analyzed in the paper; Secondly, an algorithm of table line detection based on directed unilateral connected chain is improved, which improves accuracy and speed of table line detection; Thirdly, an algorithm of constructing table framework based on hierarchical division is proposed; Finally, rapid geometry structure recognition of ballot table is implemented.(2) A ballot logic structure recognition approach based on semantic unit is proposed.Firstly, the paper proposes the concept of ballot semantic unit, then classifies the semantic units and analyzes the relationships between semantic units; Secondly, the constraints between common ballot semantic units are given; Thirdly, the logic relation tree of semantic units is constructed by recursive division; Finally, an algorithm of recognizing logic structure based on semantic unit is implemented which improves the intelligence of election system.(3) A ballot symbol recognition algorithm based on run features is proposed.To improve the accuracy of ballot symbol recognition, a ballot symbol recognition approach based on run features is proposed. Firstly, the concept of run is presented and a discriminant model is established based on run features of ballot symbols; Secondly, the relative positions of runs are described with a ternary tree; In addition, an approach of recognizing ambiguous symbols is also given; Finally, an algorithm of ballot symbol recognition based on run features is implemented, and experimental results show that the algorithm achieves high accuracy and can recognize some ambitious ballot symbols.(4) A rapid vote-counting system has been developed.A rapid vote-counting system has been developed in which the algorithms that the paper proposes are implemented. It proved the rationality and effectiveness of the methods which the paper proposed. Compared with existing vote-counting systems, it has the advantage of recognition speed and intelligence.