The problem of Optical Character Recognition (OCR) needs, as a first prerequisite, a good and powerful algorithm for binarization of scanned text. Most of the available algorithms used for binarization, uses the concept of thresholding. Such algorithms, in spite of being powerful, simple and the one with very small computational cost, are sometimes less efficient in case of scanned text of low quality. This leads to the requirement of other mathematical model for binarization. In this paper, the authors present a new mathematical model for binarization problem.