摘要：The object of research is the process of recognizing the areas of scanned documents images. The paper proposes a hierarchical approach to the segmentation of scanned documents images. This approach is an image of a scanned document in the form of a multi-level structure. At each level of the structure, images containing structural regions are highlighted. Objects of the lower level strictly correlate with a certain area of the image of the upper level: areas of the photo and graphics correspond to the image containing the illustrations, and areas of text and background to the image containing both the text and the background at the same time. Using a hierarchical approach, it is possible to perform processing separately for each image area, namely: first, the areas of illustrations are highlighted on the original image of the scanned document using the analysis of connected components. Thus, the first level of the hierarchy consists of an image containing illustrations and an image containing text with a background. Then the areas of illustrations are divided into photos and graphics by splitting the areas of illustrations into blocks, and text areas are separated from the background using processing in the neighborhood of each pixel. Thus, the second level of the hierarchy is represented by images containing homogeneous areas: photos, graphics, text and background. The hierarchical approach to segmentation has reduced the processing time by an average of 80 times. The reduction in image processing time was due to the fact that at each level and in turn, in a separate part of the hierarchical structure, it was possible to take into account the structural features of a uniform image area corresponding to this level. And also choose the signs of identification of these areas with high computational efficiency, the use of which also reduced the processing time of the scanned document.
