Exact match using PuMuPDF

[AnweshGangula]

I'm working on an app in which we need to verify if a text is an exact match (eg: "city" doesn't match "electricity").
I've looked at few existing questions( #678 & Stack-Overflow) but couldn't find an accurate solution.

I found a way that is the most accurate out of all the solutions available online (below).

But I'm facing an issue when trying to verify text in multiple lines

Here is my code:

def isExactMatch(page, term, clip):
    termLen = len(term)
    termBboxLen = max(abs(clip[0]-clip[2]),abs(clip[1]-clip[3]))
    termfontSize = termBboxLen/termLen
    f = termfontSize*2

    validate = page.get_text("blocks", clip = clip + (-f, -f, f, f), flags=0)[0][4]

    matches = len(re.findall(f'\\b{term}\\b', validate))>0

    return matches

This works perfectly fine, except in multi-line searches - I think its because the clip in page.get_text isn't spanning across multiple line

Is there any way to make it more accurate?

[JorjMcKie]

Within a text block you could check whether the exact sequence of the word(s) of your search string is equal to a sub-sequence of the block text split into words ... Just an idea.

Or extract the text as "words", sort them by block-number, line-number, vertical and horizontal coordinates. Then look wether your needle.split() matches a sub-sequence of the overall sorted word list ...

[JorjMcKie]

yeah, if, if ...
What I mean is this: suppose a PDF page like this:

And you search for needle = "for this first needle". Then this happens:

>>> rl=page.search_for(needle)
>>> len(rl)
3
>>> for r in rl:
	print(page.get_textbox(r))

	
for this
first needle
for this first needle
>>> 

My suggestions to try a solution based on words had this complication in mind.

[JorjMcKie]

It might be easier to check whether:

>>> needle.split()
['for', 'this', 'first', 'needle']
>>> # is a sublist of this list:
>>> page.get_text().split()
['I', 'am', 'looking', 'for', 'this', 'first', 'needle,', 'then', 'for', 'the', 'second', 'needle,', 'and', 'so', 'on.', 'Once', 'I', 'am', 'finished', 'looking', 'for', 'this', 'first', 'needle,', 'I', 'need', 'to', 'make', 'sure', 'that', 'I', 'am', 'not', 'computing', 'my', 'clip', 'as', 'too', 'big', 'a', 'rectangle.']

And if so how many times ...

[JorjMcKie]

Also please do read the docu for details of what may be the response of page.search_for() in the several corner cases.
You may conclude that it is not always the appropriate choice.

[AnweshGangula]

Thank you @JorjMcKie ,
Actually, what i'm trying to do is - add highlights in the PDF based on a list of needles. And before adding the highlights, checking if its an exact match.

And I found another similar discussion where you mention we can add a list of quads to highlight text in multiple lines #341 . This is what I'm looking for, but I would need to now identify which quads are from the same needle to pass them as a list.

Like in you're screenshot above, how do I combine the quads of "for this" and "first needle"

[AnweshGangula]

I think this would be a good feature to add to the page.search_for() method - to group the quads or rectangles based on the needle.

[JorjMcKie]

Like in you're screenshot above, how do I combine the quads of "for this" and "first needle"

1. Combining quads is a difficult thing - explaining this later. Combining rectangles in contrast is easy: union = rect1 | rect2.
2. Rectangle union in this case is however equal to a rectangle covering the complete first two rows! Don't know if that helps you.

Combining arbitrary quads may not make any sense, because

• a quad may be an arbitrary tetragon, i.e. not be rectangular
• two rectangular quads may not be rotated by the same angle relative to the x-axis

In both above cases it would be hard to even uniquely define what the joining quad should be. You can however easily build the envelopping rectangle of a quad (Quad.rect) and join it with other rectangles as mentioned.

Also please read the docu for how to highlight text which potentially is spread across multiple lines and you know the top-left of the begin and bottom-right of the end.

but I would need to now identify which quads are from the same needle

Where is the problem? If one of the rectangles .startswith the first word of needle you have found the start. Then check the following rectangles until you encounter one that .endswith the last word of needle.