A binarization method for degraded document image using artificial neural network and interpolation inpainting

The binarization of degraded document image is an important phase in the document analysis to extract text from the original image. For this purpose, we present a new multi-stage approach started by, a preprocessing phase to enhance the contrast of the original image and to eliminate noises. Then an interpolation inpainting is performing to create the background estimation and the foreground (text) estimation, that will be used as feature to train and test a supervised artificial neural network. the final stage is the use of the ANN to classify the pixels of the original image into 0 for black and 1 for white to get the final binary image. The suggested method has been tested and evaluated over DIBCO'2009 database, the results show the successful of the proposed approach.