Table Structure Extraction from Form Documents

Dihua Xi, Seong-Whan Lee

Abstract: Based on gradient and wavelet analyses, a novel scheme has been developed to extract table structures from skewed form document images. In this scheme, first, a skewed form document image is rotated according to the angle obtained from the gradient algorithm. Then the deskewed image is decomposed into four sub-images by divisible Multiresolution Analysis(MRA) wavelets. Afterwards, the table structure image which represents the geometric structure of the form can be obtained from the sub-images by a modified wavelet reconstruction algorithm. Meanwhile, another document image without table lines can be produced by Minkowski operation and is referred to as a table free image. Experimental results indicate that this new scheme can be applied to process the skewed form document images with promising achievements.