Thanks for the report. To determine the images associated with a page, pypdf does indeed not consider nested xobjects for image extraction.

pypdf can looks in sub XObjects, however here you are looking for an object which is part of a pattern which is not for me the way to do things.
this is a proposal to extract your image:

import pypdf

r = pypdf.PdfReader("B2.pdf")
img = pypdf.filters._xobj_to_image(r.pages[0]["/Resources"]["/Pattern"]["/P1"]["/Resources"]["/XObject"]["/X1"])[2]
img.show()

I will try to propose also a easier way to extract an image
edit. I've found a better way

with the new PR extraction will be easier:

import pypdf
r = pypdf.PdfReader("B2.pdf")
img = r.pages[0]["/Resources"]["/Pattern"]["/P1"]["/Resources"]["/XObject"]["/X1"].decode_as_image()
img.show()

Wouldn't it be better to have the fonction that should extract all images of a page to actually extract all images of the pages ?

The PDF standard said that images can be stored inside Patterns so we should expect to find images in them.

I agree that images can be stored in patterns, but the solution used inhere is not common. a pattern is expected in a context to provided a repeated image in a surface.
There is too many places where images could be (patterns, annotations, ...); will be quite complex also out of context having the image may not be very efficient.

We could implement a bool parameter recurse, deepSearch or whatever to the _page.images method.

When set to False, the standards methods _page._get_ids_image, _page._get_image would get called, keeping the image retrieval to it's simplest form, in the inline images and images dictionaries of the page.

When set to True, we could call the standard methods and return on top of their results images found in "special" cases like Patterns.

This way we still keep it efficient for the current usage.

We could implement a bool parameter recurse, deepSearch or whatever to the _page.images method.

When set to False, the standards methods _page._get_ids_image, _page._get_image would get called, keeping the image retrieval to it's simplest form, in the inline images and images dictionaries of the page.

When set to True, we could call the standard methods and return on top of their results images found in "special" cases like Patterns.

This way we still keep it efficient for the current usage.

We can propose a PR

Well well well, _page.images isn't a method but a property so passing a parameter to it isn't an option...