The article presents the results of research comparing edge detection methods in digital images and verifying their usefulness in the context of the automatic vectorization process. As part of the experiment, well-known edge detection algorithms based on the analysis of derivatives of image quality functions (Sobel, Canny, Kirch) were implemented. The research problems of the article in the case of building detection basically boil down to the identification of homogeneous areas, the detection of edges or points in a digital image. The original program developed in the Matlab environment made it possible to obtain a description of the edges and their approximation with straight lines, as well as to analyze the quality of the obtained results. In addition, the validity of using neural networks was also analyzed in this context. The neural networks used an algorithm obtained from the GitHub hosting website and implemented as a plug-in for QGIS 3.26. Another attempt at algorithmic image analysis was based on the use of the GAN technique, i.e. the use of a generative network architecture that acts as an algorithm using the potential of two mutually opposed networks whose task is to generate a synthetic result. Under this assumption, one network is the so-called data generator and the other is the discriminator, critically assessing the generating network for authenticity. For each algorithm, the accuracy of vectorization of the detected edges was calculated. The most promising in this respect was an artificial intelligence algorithm using the technique of generative adversarial networks.