diff --git a/src/ocrmypdf/optimize.py b/src/ocrmypdf/optimize.py
index 2976554c..bfe027af 100644
--- a/src/ocrmypdf/optimize.py
+++ b/src/ocrmypdf/optimize.py
@@ -111,6 +111,11 @@ def extract_image_generic(*, pike, root, log, image, xref, options):
     if pim.bits_per_component == 1:
         return None
 
+    try:
+        pim.indexed  # pikepdf 1.6.3 can't handle [/Indexed [/Array...]]
+    except NotImplementedError:
+        return None
+
     if filtdp[0] == Name.DCTDecode and options.optimize >= 2:
         # This is a simple heuristic derived from some training data, that has
         # about a 70% chance of guessing whether the JPEG is high quality,