Optimizing OCR Accuracy for Folded and Mirrored Text Documents

This literature review explores methods to enhance OCR performance on inverted, reversed, or backend text, addressing the limitations of conventional OCR systems in accurately recognizing such orientations. The methods used range from simple image processing techniques like edge detection and histogram equalization to advanced approaches such as convolutional neural networks, support vector machines, fuzzy logic, and neural networks, addressing uncertainties in character shapes, angular orientations, and feature alignments. This paper discusses the use of the TrOCR model for recognizing text on the backside of documents, addressing issues such as ink bleed-through, alongside preprocessing and postprocessing solutions. It evaluates accuracy, highlighting the model’s utility in enhancing OCR dependability. The review underscores the need for more robust algorithms capable of handling diverse document types and text orientations while preserving the originality of documents for digitization and archival purposes.