Joint blind restoration and segmentation of blurred text characters

Character segmentation of damaged printed texts is a very critic task, especially when the degradation causes the characters to touch and merge one another. With particular reference to ancient printed texts, we model the degradation as a unknown space-variant blur operator and try to jointly estimate the blur parameter and recover the undegraded image. Since the latter can be considered as a two-level image, we propose to integrate techniques of image restoration with techniques of image segmentation, based on Markov Random Field models. The problem is formulated as the minimization of a cost function which accounts for data consistency and for constraints derived from the adopted image model. A solution strategy is proposed where steps of image estimation iteratively alternate with steps of estimation for the degradation operator. To cope with the problem of space-variant blurs, we propose a recursive procedure that starting with the estimation of a single blur mask for the whole image, refines the estimate in those zones of the image where suitable validation tests, based also on a linguistic analysis, reveal an error.