Detection and localization of text in photorealistic images is a difficult, and not yet completely solved, problem. We propose the approach to solving this problem based on the method of semantic image segmentation. In this interpretation, text characters are treated as objects to be segmented. In this paper proposes the network architecture for text localization, describes the procedure for the formation of the training set, and considers the algorithm for pre-processing images, reducing the amount of processed data and simplifying the segmentation of the object “background”. The network architecture is a modification of well-known DeepLabv3 network and takes into account the specifics of images of text characters. The proposed method is able to determine the location of text characters in the images with acceptable accuracy. Experimental results of assessing the quality of text localization by the IoU criterion (Intersection over Union) showed that the obtained accuracy is sufficient for further text recognition.