How to crop each character for each line?

[nissansz]

textbox.pdf
Presentation1.pdf

For first line, I put a line of table or textbox below image,
but when I use below code to extract characters, can only get image for whole line, how to split cells?

'''
import fitz

from PIL import Image

import cv2
import numpy

path = r'L:/jp5/Presentation1.pdf'
path = r'L:/jp5/textbox.pdf'

doc = fitz.open(path)

j = 0
for i in range(0, doc.page_count):
page = doc[i]
for block in page.get_text("dict", flags=fitz.TEXTFLAGS_TEXT)["blocks"]:
for line in block["lines"]:
j = j + 1
bbox = line["bbox"] # the line bbox
text = " ".join([span["text"] for span in line["spans"]]) # text in line
if len(text.strip()) > 2:
print(bbox)
print(text)
pix = page.get_pixmap(clip=bbox, dpi=300) # pixmap of line bbox
# pix.save(...)"page%s-%s.png" % (page, xref)
# pix.save(r'F:\0mupdflines\' + "page-%i_%i.png" % (page.number, j)) # store image as a PNG

doc.close()

'''

[JorjMcKie]

You may want to use the table finder:

In [1]: import fitz
In [2]: doc=fitz.open("Presentation1.pdf")
In [3]: page = doc[0]
In [4]: tabs = page.find_tables()  # detect the tables
In [5]: tab = tabs[0]  # get first table on page
In [6]: tab.extract()  # read all cell content
Out[6]:
[['丕', '石', '右', '布', '夯', '龙', '戊'],
 ['丕', '石', '右', '布', '夯', '龙', '戊'],
 ['', None, None, None, None, None, None]]
In [7]: # also works as a pandas DataFrame:
In [8]: df = tab.to_pandas()
In [9]: df
Out[9]:
   丕     石     右     布     夯     龙     戊
0  丕     石     右     布     夯     龙     戊
1     None  None  None  None  None  None
In [10]: # you can also extract row and cell coordinates
In [11]: row = tab.rows[0]
In [12]: row.bbox
Out[12]: (0.0, 0.0, 369.3599853515625, 50.4000244140625)

In [14]: row.cells[0]
Out[14]: (0.0, 0.0, 52.768001556396484, 50.4000244140625)

For background on using PyMuPDF's table finder look at blog1 and blog2.

Hint: to make your code snippets appear as nice as mine, consider using 3 backtics (`) or use this to invoke a guide:

[nissansz]

Thank you.
How about parsing single textbox?

[JorjMcKie]

You can take each table cell and extract its content by e.g. get_textbox(cell).

[nissansz]

There is error. Also how to crop the image of the cell?

File "F:/pycharm2020.2/RapidStructure-0.0.0/exportLine_txt_png_tables.py", line 20, in
for tab in page.find_tables() :
AttributeError: 'Page' object has no attribute 'find_tables'
/
for i in range(0,doc.page_count):
page = doc[i]
# for block in page.get_text("dict", flags=fitz.TEXTFLAGS_TEXT)["blocks"]:
for tab in page.find_tables() :
a = tab.extract() # read all cell content
print(a)
/

[JorjMcKie]

install a later version

[JorjMcKie]

use page.get_pixmap(clip=cell).

[nissansz]

Cannot see cell bbox property

[JorjMcKie]

Cannot see cell bbox property

what does that mean

[nissansz]

I want to crop bbox for each character

[JorjMcKie]

sure, what's wrong with the cell bboxes?

[nissansz]

I cannot get bboxes for the cells

[JorjMcKie]

This

for cell in tab.cells:
    page.draw_rect(cell, color=(1,0,0))

gives you this:

What is missing?

[nissansz]

Thank you.
I found a problem, there should be 14 cells, but don't why the whole image is treated as an cell too.

[JorjMcKie]

The table finder tries to make sense of the vector graphics with its standard detection strategy.
Apaprt from single lines wrapping the cells, there also is large white rectangle covering the full page.
This is responsible for the "problem" taking the rest of the page (under the first 2 rows) as an additional row with only one cell.
So you have to use own code to cope with this currently. In a future ersion, we may add logic that prevents this sort of thing.

[nissansz]

Ok. Thank you.

[nissansz]

How to read a table cells by rows and columns?
There is error: cell = table.get_table_cell(row, col)
AttributeError: 'Table' object has no attribute 'get_table_cell'

'''
html = ""
cells = []
# col_widths = table.get_table_col_widths()
col_widths = table.row_count

for row in range(table.row_count):
    html += "<tr>"
    # for col in range(len(col_widths)):
    for col in range (0,col_widths):
        cell = table.get_table_cell(row, col)
        tokens = cell.get_text().split()
        bbox = cell.rect
        cells.append({
            'tokens': tokens,
            'bbox': [bbox.x0, bbox.y0, bbox.x1, bbox.y1]
        })
        html += f"<td>{' '.join(tokens)}</td>"
    html += "</tr>"

'''

[nissansz]

Thank you. But above method can only get text, not cell coordinates, any method to get coordinates by rows and columns?

[JorjMcKie]

You also have tab.cells and tab.rows.cells. These are lists of rectangles.

[nissansz]

border4.pdf

For example, abovepdf, I want to get result like below.

。作者

.....

[nissansz]

'''

。作者

.....

'''

[nissansz]

[JorjMcKie]

Try

tabs=page.find_tables()
tab=tabs[0]
for e in tab.extract():
    print(e)

    
['。作者', None, None, '件事”。“只要', '看20多份报']
['pulseawry', '认为，', '。冰壶运动需要', None, None]
['，还有师父咸鱼', '564.77', '8506.00', '', '32.20']
['Ssubs', '', '', '', '4862.66']
[None, '7', '119.57', '满了湿润和温', '68']
tabs=page.find_tables()

tab.row_count, tab.col_count
(5, 5)

For each of these cell content items you will find the corresponding cell coordinates like this

for row in tab.rows:
    print(row.cells)  # a list of rectangle tuples

When encountering joined cells, the table finder will dissolve them and alsways return the maximum values for rows / columns. It will use None cells in all the necessary places when doing this both, in terms of cell text content and cell coordinates.

for row in tab.rows:
    print([tuple(fitz.IRect(cell)) if cell!= None else None for cell in row.cells])

    
[(66, 243, 562, 271), None, None, (562, 243, 728, 300), (728, 243, 894, 300)]
[(66, 271, 231, 300), (231, 271, 397, 300), (397, 271, 562, 300), None, None]
[(66, 300, 231, 329), (231, 300, 397, 329), (397, 300, 562, 329), (562, 300, 728, 329), (728, 300, 894, 329)]
[(66, 329, 231, 387), (231, 329, 397, 358), (397, 329, 562, 358), (562, 329, 728, 358), (728, 329, 894, 358)]
[None, (231, 358, 397, 387), (397, 358, 562, 387), (562, 358, 728, 387), (728, 358, 894, 387)]

If you compare the text and the cell output you will see the logic used to dissolve the joined cells.

[nissansz]

Thank you.
It seems that None tag is difficult to identify which direction to merge, left or upper? Sometimes, it may be not clear.

[JorjMcKie]

Sure. But that's all you can have.
In your example however, it is clear.

[nissansz]

Do you know any method to judge count of merged rows/columns in pywin32 for ppt?

[JorjMcKie]

No, sorry.
Maybe the table finder can give some help though:
row_count * col_count is the count of all cells, including the None ones. That number also comes out when looking at the cells of each row.
But table.cells does not contain the None cells:

In [4]: page=doc[0]
In [5]: tab=page.find_tables()[0]
In [6]: tab.row_count * tab.col_count
Out[6]: 25
In [7]: len(tab.cells)
Out[7]: 20
In [8]:

tab.cells are exactly the visible cells. So comparing the two lists should at least be some help.