Performing OCR
==============

Use the ``aidb.perform_ocr()`` function to extract text from images. It
takes raw image bytes and a registered OCR model name as input, and
returns a table with a ``part_id`` column (identifying each text block)
and a ``text`` column (the extracted string). Because the function uses
unnesting, a single image may produce multiple rows — one per text
segment. Some providers have format restrictions; for example, the
NVIDIA NIM provider currently supports only ``png`` and ``jpeg``
formats.

**Step 1: Register an OCR-capable model**

.. code:: sql

   SELECT aidb.create_model(
       my_paddle_ocr_model,
       nim_paddle_ocr,
       credentials => {"api_key": "__NVIDIA_NIM_API_KEY__" }::JSONB
   );

**Step 2: Extract text from images**

.. code:: sql

   SELECT * FROM aidb.perform_ocr(
       decode(/9j/4AAQSkZJRgABAQAAAQABAAD/2wCEAAUFBQUFBQUGBgUICAcIC
          AsKCQkKCxEMDQwNDBEaEBMQEBMQGhcbFhUWGxcpIBwcICkvJyUnLzkzMzlHRE
          ddXX0BBQUFBQUFBQYGBQgIBwgICwoJCQoLEQwNDA0MERoQExAQExAaFxsWFRYb
          FykgHBwgKS8nJScvOTMzOUdER11dff/CABEIAWYDKgMBIgACEQEDEQH/xAAvAA
          EAAQUBAQEAAAAAAAAAAAAAAQIDBAcIBgkFAQEAAAAAAAAAAAAAAAAAAAAA/9oA
          DAMBAAIQAxAAAADsuKLhCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCR
          CRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRCRE2rpTVT
          UAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAWrtq6U1U1AAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFq7aulNVNQAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABau2rpTVTUAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAWrtq6U1U1AAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFq7aulNVNQAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAABau2rpTVTUAAAAAAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAWrtq6U1U1AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAFq7aulNVNQAAAAAAAAAAAeS/INiNdjYj8z9MAAAAAAAAAA
          FJUiQAAAAAAAAA8l60AAAAAAAAAAAAAtXbV0pqpqAAAAAAAAAAESNE617AwD5kd
          Jap+gp5L2PC/459AJ5c/ROko4DxD6EU+I4rPoY+enVJuJzfpI79ngXrU2LT88Pf
          naEcVfhHek8K9jno6fnrmH0A5c/S5YO89icId3FSOODsiOC847meN4qPoPPzx6i
          N0zw9hHeD5yeqO8Y+cfeZ6ev5uepO+nzz71OD/AKDfPnsc2BHz7xj6HPBe9AAAAA
          AAAAAALV21dKaqagAAAAAAAAAABgZ+AcB9U8rddHGO7vC9XHz43/me6NMbI3Jxwd
          YcV9C7HNYbf17pY1f0lo3uI4n3V+3+oaE745D6qNBYHPvdp89uldE78PWb15730as
          1P+h+cbw2VrDZw4F7757Nd+52Vwsd76d8b6I8H6LZ2sjWvf8AwN9Bjh7qvlbrY0P7
          bxfuDkP6Q/M/6UnPuDbzzRP6f5npD9GxuT9Y557i576EAAAAAAAAAAALV21dKaqag
          AAAAAAAAAAB+f8AoDkPrmscZeZ7zHN3ot3jgnI7uHhuRe8xwb0NuwaL5774HEXUXu
          xzt0HeHCE93DgLq7Z44Rx+9hqWxuEcQ9vRJHMHUA4O/Y7ZGpOWfoCOJMTuYci9cVD
          mPf8A+6NIei2bQfMjZn5vf5wr2z+kOReivYDgivvMeI9wAAAAAAAAAAAFq7aulNVNQ
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAiRh5gAAAAAAAAAAAAAAAWrtq6U1U1AAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFq7aulNVNQAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABau2rpTVTUAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAAAWrtq6U1U1AAAAAAAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAFq7aulNVNQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAABau2rpTVTUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAWrtq6U1U1AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFq
          7aulNVu2ZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkM
          cZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHGQ
          xxkMcZDHGQxxkMcZDHGQxxkMcZDHGQxxkMcZDHFd2xfIgAAAAAAAAAAAAAAAAAAAAAA
          AAAAAAAAAAAAAAAAAAAAAAAAAAAIrD//EADIQAAAFBAEEAQMDAgcBAAAAAAABAwQFAg
          YHERQIEDI0YBIhQBYxNxUXExgiMzZBUTX/2gAIAQEAAQgAM/2IaMaMaMaMaMaMaMaMaM
          aMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaM
          aMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaM
          aMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMaMGejIjCn+9R8kr80+1HsJfJK/
          NPtR7CXySvzT7Uewl8kr80+1HsJfJK/NPtR7CXySvzT7Uewl8kr80+1HsJfJK/NPtR7CXy
          SvzT7Uewl8kr80+1HsJfJK/NPtR7CXySvzT7Uewl8kr80+1HsJfJK/NPtR7CXySvzT7Uewl
          8kr80+1HsJfJK/NPtR7CXySvzT7Uewl+PWJq+bPtt3Q0mv7s40H92caA8uY2IvtEzEXOsUZCK
          /FMGYI/w6vsQjr6tGWlFIljTv/v8qvzT7Uewl+PV+32yLgyLyLO0y7qS6V4JjHP3dOLLBaZCu
          v8Aobk+k6AFiWehYtuM4Fv/AKgRgxVVsEezBjf3BmewRgz0NmCP7gwdX/h1aLY2exv/AMI/uDF
          VWwZiqrY6nZyZhIi0zjMSvXL7G9ouXdPYx9QIwZ6BfuDq+w+obH1D6h9Q+ob0Nj6hsGMOfzu9H
          iPqBH+TX5p9qPYS/I0LiPcFOjplL68nBdZJuiqsreWX76v+4lLesZ/jvPVutq5YsIZrf3O8/TNy
          Z3y5I2OTSFhY7Hud7uaJTJxuRsoYkuNCPuS5L4joCyFrtDGYzJmSQd1xU3B5xxeknLr4ayVXkS3
          FFHubsyubKqRgYJpYGe7sbUzNdr5ZyDjC4k4a8MgTSieOLgmIiAzpdUPa0qzPEUNll3e8RO3FlD
          Mt0zN0r2jZR4kzyuib6vFOWrzgL0ZWrdFz3JHWlBSM3Jr35ljME8sxgZayc8WEzrmDw9lx9kSM
          mIaRyVCZPim8VXe2LbfzKudmvWlAqPQzNma40blVs+1KMUZ7WQ/qJ42zFeFrXY3te8r7vBnY1tSE
          46j5TNmY3rpeMuC182Y1bFMr4OyO8yFbjs5LM+QbptfKRos6Es5Zg/x5dkyvnJ2ILppYzNUnmjND
          p28g5STy/h+XZFI2fciV22nDzidWYb7gLunq6GdjZ5vZrTMqxeSsoYluGiPuGCmWdww7CUZYc/nZ6M
          jX0yx/bjiVcsl83ZiXcPWD93mrD7xs6kMb36wyBbbeUQ/Hr80+1HsJfk3Ee4KdHTL/ACcM9SriIxj
          OKIYeynb+M05ZV5/mxt0K3SweZQZXFCdR2PZ2adR91RNn9TEtAs20XcJ3zhLLTqPSnb+sei4McvrYh
          rMv28sLSD+NcodRWPLqj14q48cwOP4ti5fWYtdbOjKbm5Zz/NfbtJERZgytb2TG0WbS2ZNaT6ZJMl
          uma02U7c0tLvf+vs6h8K4muKidcSnVLaTYq6Y5tcLy6svQU296q5RZC2bYjKOnSAaxWN457Qommqn
          WnXaOFrQsqfonYnqxMv6PZowyZHiqySMVHovvmmxrqtW/nV3xtu9VWkk0bijHuD8qz9Ekd9Wnbl4Q
          psrijMq4cxdG1QUHkbqGQu+3ZSChukv1r2GfUKHeYTQrYMGsYyaMWvVikmU9ai9ONo5tH2LaTZr1V
          oUfo6AVPAf8VQQx5ANbjzXWzdU0kRERdS1uspGw/wCrH0uy67yy5Rirhn+eXY6pZZda7ouMFsdSNqW
          xARcO1vDqLtS7rblIRx0pyaqc9cUaZfj1+afaj2EvybiPcFOjpl/k4ZphlZvGlxtkenJGyJn+sQk//bT
          H5Ee17xxknfqNqxV75zj7Euv9OO5CybEu9qi5eZ1xRbthpxsnCWdlF3a+E4K4ZOyrstTMMCu8Xv7AtgPo
          eUkGHTBKvm15ykRSszibQzc5bXCjjrHLpFNZvk55ivGqEcS8k6jpDAs6/jekoqDZ3mZz7xWMgZyQRxHbLXK
          GQXn6lj7CseCR+prPS8ZOZ7YvYvqohl3Vq29LJ9N11sZexW8IcjJM4pg6fvsU5ou/IN81xVfVgW4izRhqojx, base64),
       options => {"model": "my_paddle_ocr_model"}
   );

   __OUTPUT__
    create_model
   - -------------
   my_paddle_ocr_model
   (1 row)

    part_id |       text
   - --------+------------------
          0 | Tesseract sample
   (1 row)