Allocation of object borders is the one of the most important tasks at pavement damages recognition. They contain exhaustive information on it's form, for the subsequent analysis. The method of binarization of the initial image copes with this task. But this process is characterized by existence of a large number of distortions: washing out, gaps and loss of objects integrity, emergence of noise in homogeneous areas. Demand of the mistakes elimination has led to emergence of a large count of binarization methods. The choice of a binarization method and search of the optimum (for some set of parameters) algorithm influences the algorithms applied further to the image analysis. In this article we consider the most popular and modern binarization methods concerning problems of recognition of pavement damages.