By format we mean documents with different structure and purpose. For instance, legal documents have certain key words and “actors” that appear in certain structured format that must be strictly kept. These patterns are visually recognized by the human eye and classified accordingly with almost no effort, but automated solutions are often way too sensitive to “noise”. In this case “noise” means pieces of text that have been highlighted or handwritten that usually “confuse” the OCR and difficult the classification process. To overcome this problem it is necessary to detect these formats based on more sophisticated solutions than simply OCR recognition.
Thanks for your interest!