Use OCR to extract image text in PDFs

pyproject.toml

@@ -73,6 +73,7 @@ dependencies = [
     "gunicorn == 21.2.0",
     "lxml == 4.9.3",
     "tzdata == 2023.3",
+    "rapidocr-onnxruntime == 1.3.8"
 ]
 dynamic = ["version"]
 

src/khoj/processor/pdf/pdf_to_entries.py

@@ -68,7 +68,7 @@ class PdfToEntries(TextToEntries):
                 with open(f"{tmp_file}", "wb") as f:
                     bytes = pdf_files[pdf_file]
                     f.write(bytes)
-                loader = PyMuPDFLoader(f"{tmp_file}")
+                loader = PyMuPDFLoader(f"{tmp_file}", extract_images=True)
                 pdf_entries_per_file = [page.page_content for page in loader.load()]
                 entry_to_location_map += zip(pdf_entries_per_file, [pdf_file] * len(pdf_entries_per_file))
                 entries.extend(pdf_entries_per_file)