Optical character recognition is the subject to long-term research and development in the field of image processing. Due to wide and diverse applications of optical character recognition systems, such as in banking applications and applications that help blind people, the accuracy and reliability of those systems is very important. Therefore, in recent years more and more attention is paid to algorithms used for preprocessing the input image because their task is to adjust the input image so that further analysis enables the highest accuracy of text recognition. Preprocessing procedures implemented in this work include image enhancement and segmentation, as well as the calculation of correct text orientation and perspective in the image. The first step, image enhancement and segmentation, consists of the converting the input image to black and white using an adaptive threshold, and the application of morphological operations. The aim of their use is to improve the contours of the letters and make their shape more regular. This is followed by the algorithm that uses projective transformation for estimating and correcting the perspective of the text in the image. In this paper are described and implemented three different methods to correct vertical distortions, one of which shows excellent results for very large changes in the perspective of the original image. Applying the algorithm described in this paper to the input image, before the procedure for character recognition, OCR accuracy increases by 100%.