Discard cropped / hidden content ("sanitize") or extract cropped images instead of raw images

[quanvinh]

My PDF has tons of cropped images and AFAIK PyMuPDF only allows me to extract raw (uncropped) ones.

I was wondering if any of the following is possible?

a. Discard hidden / cropped part of all images (similar to "Redact" -> "Sanitize" in Acrobat, without rasterizing) prior to extracting.
b. Obtain cropbox of each images so I can crop the extracted raw images using another library.
c. (Preferrably) Ignore cropped data during extracting (aka extract just the cropped images instead of raw ones).

[JorjMcKie]

You need to know which part of the image is covered by other stuff. This would be a sequence of rectangles which have a non-empty intersection with the image bbox on the page.
Then use the image's transformation matrix to compute those intersections in original image coordinates. For all those intersections, empty the resp. rectangle in the image pixmap (i.e. set it to white).
Method page.get_image_bbox(item, transform=True) will return that transforation matrix alongside the image bbox.
Then look at this for details on transformation matrix.

• Let pix be the pixmap of the extracted image. Then we define matrix expand = fitz.Matrix(pix.width, 0, 0, pix.height, 0, 0).
• Let intersect be an intersection area on the page in page coordinates.
• Then xrect = (intersect * ~transform * expand).irect is an area in the pixmap we need to blank out.
• Depending in the pixmap colorspace (GRAY, RGB, CMYK, with or without alpha), we use the correct pixel value for "white" and execute pix.set_rect(xrect, white). If you have for example an RGB image without alpha, you would use white = (255, 255, 255).

[JorjMcKie]

Also note, that there is Pixmap.copy(sourcepix, irect), which lets you make sub-pixmaps of a given source pixmap only containing the irect part.