Analysis and Benchmarking of OCR Accuracy for Data Extraction Models