Novel approach to background-text-nontext separation in ancient degraded document images

Nowadays lots of handwritten and printed ancient documents needs to perform to digitized form for automated processing and analysis. In this paper, an approach to background-text-nontext separation procedure based on differences of presented in a document image objects sizes which can be obtained by binarization and segmentation algorithms, is proposed. After binarization by a proper method, it is segmented and the distribution of segments sizes is obtained. It is assumed that the three types of objects presented in an image have significantly different sizes; therefore, the problem of separation comes to discrimination of the set of segments into three groups. The thresholds for the separation of these groups can be found by minimizing the intrasample variation which is used in discriminant analysis. Some examples of images from the Matenadaran collection are considered and the separated parts of the image are illustrated and interpreted.