自适应版面分割的非线性随机滤波方法

版面分割是文本分析领域中的一个重要而基础的研究课题，传统的版面分割方法主要包括两大类，即top-down和bottom-up方法，这两种方法虽然沿用至今，但它们对处理具有高几何复杂性的文本并不是十分有效，此外，分割文本的过程需要反复迭代，因而耗费计算时间，近年提出的修正分形特征（MFS）方法虽能克服top-down和bottom-up方法的上述缺点，但由于需要计算修正分形特征，因此无论从原理的简明扼要，还是降低算法复杂度等方面来考虑，都有进一步改进的必要，作者提出了基于次序统计滤波器（OSF）的版面分割新方法，它的原理较之MFS方法更加直观和朴素，在工程实现上它不仅能像MFS方法那样自适应地处理各种具有高几何复杂度的文本，而且更加节省计算时间。

NONLINEAR STOCHASTIC FILTERING METHODS OF ADAPTIVE PAGE SEGMENTATION

Page segmentation is one of the important and basic research subjects of document analysis. Traditionally, there are two major kinds of page segmentation approaches, i.e., top down approach and bottom up approach. Though these two approaches are been used up to now, they are not effective for processing documents with high geometrical complexity and the process of splitting document needs iterative operations which is time consuming. The Modified Fractal Signature (MFS) approach which was presented in recent years can overcome the above weaknesses, however it needs to calculate modified fractal signature. Hence no matter the simplicity of the theory and the complexity of the algorithm, there is a need to make further improvement. The authors shall present a new approach for page segmentation based on order statistic filter (OSF). Comparing the theory of this approach to MFS approach, it is more direct and simpler. In implementation, it can not only adaptively process the documents with high geometrical complexity like MFS approach, but also save a lot of time of computing.