The goal of our research is to develop a set of methods and algorithms for automatic data extraction from images of historical astrographic plates. We focus on scanned astrographic maps that are result of the 19th century huge astronomical project Carte du Ciel. The goal of Carte du Ciel, along with Astrographic Catalogue, was to map the stars up to 14th magnitude in the entire visible sky. The result is stored in the form of photographic plates and their paper copies, called astrographic maps. They contain valuable information for researchers in the field of astronomy, mainly because of the age of the data, and are a subject of extensive research. In this paper we present our approach for segmentation of images of astrographic maps and automatic extraction of the triple stars images (called asterisms) that are contained in them. The presented method creates a data set of asterisms whose purpose is to train a convolutional neural network for automatic flare stars detection in the astrographic maps.