PDFMiner seems to be unable to extract images from scanned PDFs

[pettzilla1]

Hi,

I'm trying to read the images off a scanned pdf to conduct OCR on them , I can't include this PDF for data privacy reasons.

General code run -

def _parse_lt_objs(lt_objs):
    for lt_obj in lt_objs:
        if isinstance(lt_obj, LTImage):
           iw = ImageWriter('output_dir')
           iw.export_image(lt_obj)
        elif isinstance(lt_obj, LTFigure):
            image_text+= _parse_lt_objs(lt_obj, page_number)

pdfrsrcmgr = PDFResourceManager()
pdflaparams = LAParams()
pdfdevice = PDFPageAggregator(pdfrsrcmgr, laparams=pdflaparams)
pdfinterpreter = PDFPageInterpreter(pdfrsrcmgr, pdfdevice)
img_content = []
fp = open(file, 'rb')
print(PDFPage.create_pages(fp))
i=0
for page in PDFPage.get_pages(fp):
    pdfinterpreter.process_page(page)
    # receive the LTPage object for this page
    pdflayout = pdfdevice.get_result()
    img_content.append(_parse_lt_objs(pdflayout))
    i+=1
fp.close()

---

Error given
ValueError: Unsupported `bitspercomponent': 1

Hi,
It is possible to have one bit per component PNGs especially with scanned pdfs

in the section https://github.com/pdfminer/pdfminer.six/blob/master/pdfminer/utils.py

please look at the page here https://stackoverflow.com/questions/757265/how-does-pdfs-bitspercomponent-translate-to-bits-per-pixel-for-images

some work needs to be performed to insure correct images can be extracted

[pietermarsman]

The [PNG specification] only mentions bytes, not bits. The apply_png_predictor() needs a lot of rework when it needs to be applied on bits instead of bytes. But it might be easier if your PNG does not use a filter, which does make sense if it only uses bits. Can you check the filter type?

[pettzilla1]

Hi Pietermarsman,
I removed the element that throws the error and it successful saved the image to .btm while this was not readable when I used pillow to convert it to JPG it could now be displayed , it seems that the code around line 136 of "utils.py " should be amended as a bits per component of 1 is supported.

[pettzilla1]

this fix worked for me @pietermarsman , line 135 in util.py should be " if bitspercomponent not in [1,8]:"