From 09c4c1f86757ebb1c650a418768c5c5577a0ba74 Mon Sep 17 00:00:00 2001
From: Philippe Prados <github@prados.fr>
Date: Wed, 26 Mar 2025 15:01:16 +0100
Subject: [PATCH] Fix images parser

---
 .../document_loaders/parsers/images.py        | 31 ++++++++++---------
 1 file changed, 16 insertions(+), 15 deletions(-)

diff --git a/libs/community/langchain_community/document_loaders/parsers/images.py b/libs/community/langchain_community/document_loaders/parsers/images.py
index e8df541291f..3d977aae973 100644
--- a/libs/community/langchain_community/document_loaders/parsers/images.py
+++ b/libs/community/langchain_community/document_loaders/parsers/images.py
@@ -45,27 +45,28 @@ class BaseImageBlobParser(BaseBlobParser):
         """
         try:
             from PIL import Image as Img
-
-            with blob.as_bytes_io() as buf:
-                if blob.mimetype == "application/x-npy":
-                    try:
-                        img = Img.fromarray(numpy.load(buf))
-                    except EOFError:
-                        return  # Ignore too small images
-                else:
-                    img = Img.open(buf)
-                content = self._analyze_image(img)
-                logger.debug("Image text: %s", content.replace("\n", "\\n"))
-                yield Document(
-                    page_content=content,
-                    metadata={**blob.metadata, **{"source": blob.source}},
-                )
         except ImportError:
             raise ImportError(
                 "`Pillow` package not found, please install it with "
                 "`pip install Pillow`"
             )
 
+        with blob.as_bytes_io() as buf:
+            if blob.mimetype == "application/x-npy":
+                array = numpy.load(buf)
+                if array.ndim == 3 and array.shape[2] == 1:  # Grayscale image
+                    img = Img.fromarray(numpy.squeeze(array, axis=2), mode="L")
+                else:
+                    img = Img.fromarray(array)
+            else:
+                img = Img.open(buf)
+            content = self._analyze_image(img)
+            logger.debug("Image text: %s", content.replace("\n", "\\n"))
+            yield Document(
+                page_content=content,
+                metadata={**blob.metadata, **{"source": blob.source}},
+            )
+
 
 class RapidOCRBlobParser(BaseImageBlobParser):
     """Parser for extracting text from images using the RapidOCR library.