How to extract image

[pranjal-jaiswal]

How to extract the images of the pdf?

[samkit-jain]

Hi @pranjal-jaiswal Appreciate your interest in the library. You can use the .images property to extract the images in a page of a PDF.

import pdfplumber

pdf = pdfplumber.open("file.pdf")

for page in pdf.pages:
    for image in page.images:
        do_something(image)

[pranjal-jaiswal]

Hi @samkit-jain, Thanks for the prompt reply and help. But .images give list of dictionary object with details of the image. I have attached a sample bellow. I want to save these images and process OCR on them. Please help me in this if you can. I know one method of cropping the image out of the page but I want a better solution. Thanks again for your help.

{'x0': Decimal('438.420'), 'y0': Decimal('104.640'), 'x1': Decimal('776.580'), 'y1': Decimal('507.360'), 'width': Decimal('338.160'), 'height': Decimal('402.720'), 'name': 'Im0', 'stream': <PDFStream(398): raw=48424, {'Subtype': /'Image', 'Length': 48423, 'Filter': /'DCTDecode', 'BitsPerComponent': 8, 'ColorSpace': PDFObjRef:378, 'Width': 500, 'Height': 595, 'Interpolate': True, 'Type': /'XObject'}>, 'srcsize': (Decimal('500'), Decimal('595')), 'imagemask': None, 'bits': 8, 'colorspace': [[/'ICCBased', <PDFStream(397): raw=2599, {'Length': 2598, 'Filter': /'FlateDecode', 'N': 3, 'Alternate': /'DeviceRGB'}>]], 'object_type': 'image', 'page_number': 1, 'top': Decimal('104.640'), 'bottom': Decimal('507.360'), 'doctop': Decimal('104.640')}

[jsvine]

Hi @pranjal-jaiswal, unfortunately pdfplumber does not currently provide a method for extracting the images embedded in a PDF. You might try working with the pdfminer object directly, via pdf.doc; see #456 (comment) for details.

[samkit-jain]

You can use something similar to the following.

from pdfminer.image import ImageWriter
from pdfminer.layout import LTImage

imagewriter = ImageWriter("/path/to/output/directory/")

for p in pdf.pages:
    for obj in p.layout:
        if isinstance(obj, LTImage):
            imagewriter.export_image(obj)

A word of caution though that so far I have been unable to extract LTImage objects. Like @jsvine referenced, you can try using the PDFDocument object and see if you are able to extract the LTImage objects in the PDF. You can also use the CLI tool pdfimages for the same.

[pranjal-jaiswal]

Thanks a lot @samkit-jain and @jsvine for your help.

[luojunhui1]

i still have this problem in 2023, is there any efficient or recommended methods for me to extract the images in PDF? the advice of @samkit-jain enlightens me to check the code of pdfminer, however, i can't find the way to transfrom the dict like

'Height': 686, 'ColorSpace': /'DeviceRGB', 'BitsPerComponent': 8, 'Interpolate': False, 'SMask': <PDFObjRef:55>, 'Filter': /'FlateDecode', 'Length': 33238}>, 'srcsize': (1234, 686), 'imagemask': None, 'bits': 8, 'colorspace': [/'DeviceRGB'], 'object_type': 'image', 'page_number': 7, 'top': 228.37, 'bottom': 451.21999999999997, 'doctop': 5279.889999999999}

to a LTImage object, could you give me any advice, thanks a lot.

[LiBinNLP]

def extract_pdf_images(cur_page, image_dir):
    images = cur_page.images
    for idx, image in enumerate(images):
        bbox = [image['x0'], cur_page.cropbox[3]-image['y1'],  image['x1'], cur_page.cropbox[3]-image['y0']]
        img_page = cur_page.crop(bbox=bbox)
        img_obj = img_page.to_image(resolution=500)
        page_number = image['page_number']
        image_name_prefix = f'{page_number}-{idx + 1}'
        image_name = f'{image_name_prefix}' + ".png"
        image_path = f'{image_dir}/{image_name}'
        img_obj.save(image_path)

[samanth-jain]

import pdfplumber
import io
from PIL import Image
from IPython.display import display

pdf = pdfplumber.open("sample-ww2.pdf")
all_images = []
for page in pdf.pages:
    for image in page.images:
        all_images.append(image) 

for img in all_images:
    image_dict = img
    image_stream = image_dict['stream']
    image_data = image_stream.get_data()
    image_bytes = io.BytesIO(image_data)
    real_image = Image.open(image_bytes)

    display(real_image)