E. Badekas and N. Papamarkos


Binarization, thresholding, document processing, detector parameters, evaluation


Most of the document binarization techniques have many parameters that must initially be specified. Usually, document binarization evaluation is subjective and employs human observers for the estimation of the proper set of parameters, for each one of the binarization techniques. The selection of the proper values for these parameters is crucial for the final binarization result. However, there is no set of parameters that guarantees the best binarization result for all document images. Thus, the proper parameters must be adapted to each one of the document images. This paper proposes a new technique which permits the estimation of the proper parameters values for each one of the document binarization techniques. The proposed approach is based on a statistical performance analysis of a set of binarization results obtained by the application of a binarization technique, using different parameters values. Using the statistical performance analysis, the best document binarization result of a set of document binarization techniques can also be estimated. From several experimental results, we verify that the proposed evaluation technique is successfully applied to different types of document images. In addition, psycho-visual experiments show that the selection of the best binarization technique, obtained by the proposed approach, agrees in most of the cases, with the human perception ability.

Important Links:

Go Back