Want to pass CV2 png in ram to pdf stream, how to do it?

[nissansz]

Below code is fine to write to local disk.
params = [cv2.IMWRITE_PNG_COMPRESSION, 9]
cv2.imencode('.png', new, params)[1].tofile( target)

But, I want to use it directly in ram, instead of writing to local file, how to insert above compressed result into pdf ?
page.insert_image(page.rect, stream=fp.getvalue())

import fitz
import io
from PIL import Image

...

img = doc.extractImage(xref)
bio_in = io.BytesIO(img["image"]) # just to create a file-like object
pil = Image.open(bio_in) # to be used for Pillow
bio_out = io.BytesIO() # the output file-like
pil.save(bio_out, format="jpeg") # write to it in a space-saving format
imgdoc = fitz.open("jpg", bio_out.getvalue()) # open it as a PyMuPDF document
pdfbytes = imgdoc.convertToPDF() # return the PDF image of that

... and so on like in one of the previous posts

[JorjMcKie]

I have no experience with CV2. You must find a way to let it write to a general Python file object - as opposed to a filename string.
You then can prepare a file object via target = io.BytesIO() and tell CV2 to write to target.
Then use the bytes object target.getvalue() further on.

[nissansz]

There seems no way to do that.
Can only use jpeg, which seems larger than original pdf file size. And clarity may be lower.

[JorjMcKie]

Well, blame CV2 then.

[nissansz]

Blame Pillow

[JorjMcKie]

disagree: a package outputting stuff that is unable to write to memory is decades behind

[nissansz]

is_success, im_buf_arr = cv2.imencode('.png', imgsrc, params)
fp = io.BytesIO(im_buf_arr)

above code can save to IO,
but after inserting this stream to pdf, it is same result as pillow, super large size. Compression did not work when using io.

[JorjMcKie]

ridiculous package!

[JorjMcKie]

you might consider to reprocess (copy) the io output to another io with Pillow, this time using "optimize=True".

[nissansz]

can you help try to test convert png from io.BytesIO() into pdf? To see whether the file size becomes super large, the only solution seems to use jpeg, which is not as good as png

new_image.save(fp, "JPEG") # saving the image in memory

[JorjMcKie]

you are using garbage=3,deflate=True?

[JorjMcKie]

The following delivers 2 files of identical size:

import fitz,pathlib
doc=fitz.open()
page=doc.new_page()
imgfile="nur-ruhig.jpg"
page.insert_image(page.rect,filename=imgfile)
5
doc.ez_save("file.pdf")
doc.close()
doc=fitz.open()
page=doc.new_page()
page.insert_image(page.rect,stream=pathlib.Path(imgfile).read_bytes())
5
doc.ez_save("stream.pdf")

[nissansz]

No option used. see below code

            page = output_doc.new_page(width=595,
                                       height=840)  # make page in target Document
            page.insert_image(page.rect, stream=fp.getvalue())

    # doc.delete_pages(0, 1)
    # output_doc.insert_pdf(doc) 

    # output_doc.save(file+"output_file.pdf")
    output_doc.save(pdf)
    output_doc.close()
    doc.close()

[JorjMcKie]

Well, give PyMuPDF a chance to compress the stuff by all means!
Reading the documentation, you will find the recommendation to do that, especially for PNG, which may contain an alpha channel.

[nissansz]

Any sample code to compress from io?

…

------------------ 原始邮件 ------------------ 发件人: "Jorj X. McKie" ***@***.***> 发送时间: 2022-12-06 21:30:57 收件人: pymupdf/PyMuPDF ***@***.***> 抄送: nissansz ***@***.***>,Author ***@***.***> 主题: Re: [pymupdf/PyMuPDF] Want to pass CV2 png in ram to pdf stream, howto do it? (Discussion #2106) Well, give PyMuPDF a chance to compress the stuff by all means! Reading the documentation, you will find the recommendation to do that, especially for PNG, which may contain an alpha channel. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[JorjMcKie]

What has io to do with all that?
We are talking about the Document.save() parameters.

[nissansz]

tried below, size is large than jpeg 95

output_doc.save(pdf+'_replaced_png.pdf', garbage=3,deflate=True)

[JorjMcKie]

but smaller than naked save()?
anyway that is all you can achieve

[nissansz]

The strange thing is that

1. first processing with local png files fold, jpeg default 75, pdf size is 3MB.
2. second processing with above pdf to make slight image rect replacement, jpeg default 75, pdf size became 5MB.

[nissansz]

I found below the problem may be due to the first step to read image from pdf. After below reading, the image became large than original png before combining pdf. After I read page from combined pdf, then each image became larger, so no matter how I compress, the image still is large from the beginning from the reading step, any method to read image in original size?

for i in range(0, doc.page_count):
    zoom = 3.5
    mat = fitz.Matrix(zoom, zoom)
    page = doc[i]
    # pix = orig_page.get_pixmap(matrix=mat)
    pix = page.get_pixmap(matrix=mat)
    img = Image.frombytes("RGB", \
                          [pix.width, pix.height], \
                          pix.samples)
    imgsrc = cv2.cvtColor(numpy.asarray(img), cv2.COLOR_RGB2BGR)