Lost in OCR Translation? Vision-Based Approaches to Robust Document Retrieval