how can i get the sub image in every page image?

[xiongzl]

i want to get the pos of image in every pdf img, i tried to use the obj type ,but get the wrong pos.

def get_page_svg(page,scale):
    svg_box = []
    page_width = page.get_width()
    page_height = page.get_height()
    # objects = page.get_objects()
    objects = page.get_objects(filter=[pdfium.raw.FPDF_PAGEOBJ_IMAGE], max_depth=50)
    for index , obj in enumerate(objects):
        left, bottom, right, top = obj.get_pos()
        y_top_cv = page_height - top
        y_bottom_cv = page_height - bottom
        x1, y1 = int(left*scale), int(y_top_cv*scale)
        x2, y2 = int(right*scale), int(y_bottom_cv*scale)
        if obj.type in [2,3,4,5] :
            print('obj.type : ',obj.type)
            img_pos = {
                "x": x1,
                "y": y1,
                "width": x2-x1,
                "height": y2-y1,
            }
            if x2 -x1 > 15 and y2-y1 > 15:
                svg_box.append(img_pos)
    return svg_box

i wonder what this different obj.type specifically correspond to.

is this?
ObjectTypeToStr= {0: '?', 1: 'text', 2: 'path', 3: 'image', 4: 'shading', 5: 'form'}

docs screenshot (click to show)

edits by mara004: formatting fixes, highlighting, hid docs screenshot by default

[mara004]

Uhm, I would hazard a guess that these particular images might actually be nested in a Form XObject container (you can basically imagine this like another page nested in the page). Thus, coordinates of sub-objects are relative to the Form XObject itself, not to the main page.
Try highlighting objects of type FPDF_PAGEOBJ_FORM to check if that hypothesis is correct.

As far as I know, upstream pdfium does not have a public API that would allow to translate Form XObject coordinates to the main page coordinates, so you can only work around this issue by setting max_depth=1 to exclude images from within Form XObjects altogether.

I've been aware of this problem for a while, as it affects post-processing in pypdfium2's built-in rendering CLI as well:

pypdfium2/src/pypdfium2/_cli/render.py

Line 307 in f2c3a54

# FIXME pdfium does not seem to provide APIs to translate XObject to page coordinates, so not sure how to handle images nested in XObjects.