A novel form detection and removal scheme for document images
Tien-Ying Kuo, Yi-Chung Lo · 2010
In this work, we propose a novel algorithm to extract form structure from form documents. First, an effective image binarization method is proposed as a preprocessor for the purpose of color quantization, which is able to deal with complex images by adaptively taking the global and local image characteristics into account. Then, a form feature extraction unit modeled by Markov random field (MRF) is followed to extract the form feature. It estimates the probability of each pixel from its neighborhood state, and recognizes the form structure based on the probability result. Unlike the geometric-based method in literature, our probability-based form feature extraction algorithm is more effective and can successfully recognize any size of form cells. In the experiment, our method has been integrated into a digital archiving system to evaluate its effectiveness in terms of both the visual and f-measurement numerical performance.