Processing Algorithm of Irregular Table Image in Tobacco Package Based on Dual-coding Difference of Gaussians Method

Abstract

Objectives: Currently, in a large number of print-out report documents from tobacco package, there exist irregular phenomena such as discontinuous vertical lines, misplaced frame lines and multi-page tables. Thus, the existing table recognition algorithm cannot be adopted to perform digital identification. In order to solve this problem, this paper proposes a table image processing algorithm based on the dual-coding difference of Gaussians iterative clustering. Firstly, the method of local regional sub-block is used to the skew correction threshold to conduct image correction. Secondly, the corrected images are coded by rows and columns, and 2D image features are transformed into 1D image features. Thirdly, the Gaussian differenced operation is adopted to obtain effective characteristic matrices that are stable and easily distinguishable. Then the iterative clustering analysis is performed to obtain the feature values of effective frame lines. Fourthly, after finishing the tasks, such as the table positioning, inner structure reconstruction, and text information identification, the dichotomy judgmentsof the integrity of multi-page tablesare realized according to the local pixel features. Finally, the text information inside the local regions and the reconstructed regions are merged, and the digital reproduction of the multi-page tables is realized. To validate the effectiveness of the proposed algorithm, an experiment in the sample set containing 12,840 table images with different resolutionsis carried out. The average detection accuracies of table positioning, table cell reconstructionand multi-page incompleteness are 98.95%, 99.80%, and 95.85%, respectively. The experimental results show that the proposed algorithm is simple and effective, and can accomplish the digital reproduction of irregular tables.