Character segmentation in highly blurred ancient printed documents

Character segmentation, which is a fundamental preliminary step for character recognition, is particularly critical in the case of ancient printed documents, where several degradation processes may cause the characters to touch and merge one another. In the paper the problem of character segmentation is formulated as the minimization of a cost function which accounts for data consistency and expresses both radiometric constraints on the image gray levels and geometrical constraints on the character boundaries. Since the degradation operator, which derives from several and diverse factors, is also unknown, it must be estimated along with the segmented image. Hence, we propose a solution strategy where steps of image estimation iteratively alternate with steps of estimation for the degradation operator. Results of both simulated and real experiments are shown to validate the method