Research on Automatic Correction of the Document Images Based on Perspective Transformation
Biao Jin, Yu‐xin Liu, Dan Liu, Wenfa Qi, Yajun Chen, Shaokang Wang · 2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI) · 2021
With the rapid popularization of mobile terminals and prompt development of information technology, a large number of document images will be captured through smart phone devices. Due to the differences in hardware performance and photographing methods, the document images usually have some distortions, such as tilt, shear, trapezoidal deformation. To improve the document image utilization efficiency, this paper proposes an automatic document image correction method based on perspective transformation. Firstly, the positions of all text lines are located by image morphological transformation. Secondly, the text image region to be corrected is detected by the relative position offset of the text lines. Then, the left and right boundary pixel sets of all text lines in the text image region are extracted respectively, and the corresponding boundary lines are fitted by the least square method. Finally, the minimum circumscribed quadrilateral of the text image region is determined according to the fitted left and right boundary lines along with top and bottom text line boundary lines, and the corrected document image is obtained by image perspective transformation. The experimental results show that the proposed method can accurately perform automatic correction of the images taken from paper-based documents, which has higher computational efficiency and stronger practicability