Texts contained in digital images are usually limited in terms of interaction and editing by users that need to extract this content. Thus, this article proposes an Optical Character Recognition system capable to convert a text contained in the image to an editable text file. For segmentation step, the Connected Contours method were more efficient, with accuracy greater than 99% and processing time less than one second for all possibilities tested. To pattern recognition step, the extractor Central Moments with Multi-Layer Perceptron network were the best combination for the system developed with 99.86% accuracy, 99.93% specificity and testing time in the order of nanosecond.