Document image binarisation based on low-rank modelling

Segmentation from document images with a complex background is presented. The background component is estimated by low-rank modelling. With the estimated background component, the original image is further transformed into a compensated image which is easily segmented with Otsu´s thresholding algorithm to obtain a good binarisation result. Experiments on the Bickley diary and the DIBCO 2013 datasets validate the feasibility of the proposed method.