Skip to content

Commit 189a43e

Browse files
JorjMcKieJorjMcKie
committed
Document revised OCR processing
1 parent c3d4e89 commit 189a43e

3 files changed

Lines changed: 147 additions & 6 deletions

File tree

‎docs/pymupdf4llm/api.rst‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -85,13 +85,11 @@ The PyMuPDF4LLM API
8585
:arg bool footer: |PyMuPDFLayoutMode_Valid| boolean to switch on/off page footer content. This parameter controls whether to include or omit footer text from all the document pages. Useful if the document has repetitive footer content which doesn't add any value to the overall extraction data. Default is `True` meaning that footer content will be considered.
8686

8787
:arg bool force_ocr: |PyMuPDFLayoutMode_Valid| if `True`, OCR will be applied to all pages regardless of their content.
88-
89-
Only works when `ocr_function` is specified.
90-
88+
9189
This may be useful for documents which are known to be image-based and thus profit from OCR, but which do not meet the default criteria for applying OCR. Default is `False` meaning that OCR will only be applied to pages which meet the default criteria.
9290

9391
.. warning::
94-
Requires `ocr_function` to be specified otherwise an exception will be raised.
92+
Requires that either one of the default supported OCR engines is installed or `ocr_function` specifies a callable OCR function. Otherwise, an exception will be raised.
9593

9694
:arg bool force_text: generate text output even when overlapping images / graphics. This text then appears after the respective image.
9795

@@ -121,9 +119,9 @@ The PyMuPDF4LLM API
121119
* `(top, bottom)` yields `(0, top, 0, bottom)`.
122120
* To always read full pages **(default)**, use `margins=0`.
123121

124-
:arg int ocr_dpi: |PyMuPDFLayoutMode_Valid| specify the desired image resolution in dots per inch for applying OCR to the intermediate image of the page. Default value is 300. Only relevant if the page has been determined to profit from OCR (no or few text, most of the page covered by images or character-like vectors, etc.). Large values may increase the OCR precision but increase memory requirements and processing time. There also is a risk of over-sharpening the image which may decrease OCR precision. So the default value should probably be sufficiently high.
122+
:arg int ocr_dpi: |PyMuPDFLayoutMode_Valid| specify the desired image resolution in dots per inch for applying OCR to the intermediate image of the page. Default value is 300. Only relevant if the page has been determined to profit from OCR (no or few text, most of the page covered by images or character-like vectors, etc.). Larger values do not usually increase the OCR precision. There also is a risk of over-sharpening the image which may decrease OCR precision. So the default value should probably be sufficiently high - in many cases you should see satisfactory results already with values of 150 or 200. Be aware that this value has a O(n²) impact on processing time and memory requirements.
125123

126-
:arg callable ocr_function: |PyMuPDFLayoutMode_Valid| if you want to provide your own :ref:`OCR function <pymupdf_layout_ocr_engines>`, specify it here. If omitted (`None`), the built-in Tesseract OCR engine will be used.
124+
:arg callable ocr_function: |PyMuPDFLayoutMode_Valid| if you want to provide your own :ref:`OCR function <pymupdf_layout_ocr_engines>`, specify it here. If omitted (`None`), one of the available built-in OCR engines will be used.
127125

128126
:arg str ocr_language: |PyMuPDFLayoutMode_Valid| specify the language to be used by the Tesseract OCR engine. Default is "eng" (English). Make sure that the respective language data files are installed. Remember to use correct Tesseract language codes. Multiple languages can be specified by concatenating the respective codes with a plus sign "+", for example "eng+deu" for English and German.
129127

‎docs/pymupdf4llm/index.rst‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -32,6 +32,7 @@ Features
3232
- Support for image and vector graphics extraction.
3333
- Layout analysis for better semantic understanding of document structure.
3434
- Support for page chunking output.
35+
- Automatic detection of pages which profit from OCR and support for various OCR engines.
3536
- Integration with :ref:`LlamaIndex <integration_with_llamaindex>` & :ref:`LangChain <integration_with_langchain>`.
3637

3738
API

‎docs/pymupdf4llm/ocr-plugins.rst‎

Lines changed: 142 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,142 @@
1+
Default OCR Functions
2+
======================
3+
4+
PyMuPDF4LLM supports default OCR functions. They come in the form of plugins that are present in its `ocr` subpackage. They are based on currently 3 popular OCR engines, Tesseract OCR, RapidOCR and PaddleOCR. Some engines can be combined to make use of their strengths and mitigate their weaknesses. For example, Tesseract OCR is very good at **recognizing** text, while RapidOCR is better at **detecting** text bounding boxes in images with complex backgrounds. By combining the two engines, we can achieve better overall OCR results while at the samne time also reducing the overall OCR processing time.
5+
6+
Here is an overview of the available default plugins:
7+
8+
============== ========================= =================================================================================
9+
Plugin Name Engines Description
10+
============== ========================= =================================================================================
11+
rapidocr_api RapidOCR Uses RapidOCR for both text **detection** and text **recognition**
12+
paddleocr_api PaddleOCR Uses PaddleOCR for both text **detection** and text **recognition**
13+
tesseract_api Tesseract OCR Uses Tesseract OCR for both text **detection** and text **recognition**
14+
rapidtess_api RapidOCR + Tesseract OCR Uses RapidOCR for text **detection** and Tesseract OCR for text **recognition**
15+
paddletess_api PaddleOCR + Tesseract OCR Uses PaddleOCR for text **detection** and Tesseract OCR for text **recognition**
16+
============== ========================= =================================================================================
17+
18+
If not explicitely selected via the `ocr_function` parameter, PyMuPDF4LLM will check the availability of the three OCR engines and pick one of the above plugins in the following order of preference:
19+
20+
1. `rapidtess_api` (if both RapidOCR and Tesseract OCR are available)
21+
2. `paddletess_api` (if both PaddleOCR and Tesseract OCR are available)
22+
3. `rapidocr_api` (if RapidOCR is available, but not Tesseract OCR)
23+
4. `paddleocr_api` (if PaddleOCR is available, but not Tesseract OCR)
24+
5. `tesseract_api` (if Tesseract OCR is available, but neither RapidOCR nor PaddleOCR are available)
25+
26+
If none of these engines is available (and no own plugin is provided), no OCR will be performed at all. If the `force_ocr` parameter is ``True``, an error will be raised. Otherwise, the document will be processed without OCR and a warning will be displayed.
27+
28+
The chosen plugin is displayed as an information message.
29+
30+
How Default Plugins Work
31+
------------------------
32+
33+
The provided default plugins use the following **"hybrid"** OCR approach:
34+
35+
1. Each page is cleaned from any existing standard text content.
36+
2. The remaining page is rendered as an image and passed to the OCR engine for text detection and recognition.
37+
3. Only the detected text is inserted back into the original page as standard text content.
38+
39+
In this way, all original content (text and other elements) is preserved and only **augmented** with the newly recognized text. This allows for a more accurate and complete text extraction while also preserving the original document structure and formatting as much as possible. It also allows for a more efficient OCR processing since only the non-extractable text is processed by the OCR engine. This can significantly reduce the overall processing time.
40+
41+
It also increases the chances for a successful layout detection, because other original content like vectors remains intact and is not rendered to pixels.
42+
43+
Forcing the Choice of a Default Plugin
44+
---------------------------------------
45+
The default plugins are designed to be used as is, without any need for configuration. However, if you want to use a specific plugin, you can do so by using the following approach (which enforces for instance using RapidOCR and skipping above selection process). Please note that all plugins have a function named `exec_ocr` that does the actual OCR::
46+
47+
import pymupdf4llm
48+
from pymupdf4llm.ocr import rapidocr_api
49+
50+
my_ocr_function = rapidocr_api.exec_ocr
51+
52+
# Use my_ocr_function as the OCR function in PyMuPDF4LLM
53+
md_text = pymupdf4llm.to_markdown("input.pdf", ocr_function=my_ocr_function)
54+
55+
56+
Providing your Own Plugin
57+
-------------------------
58+
59+
If you want to use your own OCR function, you can do so as follows::
60+
61+
import pymupdf4llm
62+
63+
def my_ocr_function(page, pixmap=None, dpi=300, language="eng"):
64+
# Your OCR implementation here
65+
return None
66+
67+
# Use my_ocr_function as the OCR function in PyMuPDF4LLM
68+
md_text = pymupdf4llm.to_markdown("input.pdf", ocr_function=my_ocr_function)
69+
70+
Your plugin must accept at least the ``page`` parameter which is a PyMuPDF Page object. The other parameters are optional. The plugin must create (or extend) the text of the passed-in page object by simply inserting text (using any of PyMuPDF's text insertion methods). No return values expected.
71+
72+
Be prepared to accept ``None`` or a PyMuPDF Pixmap object as the `pixmap` parameter, which is the rendered image of the page if provided. Parameters ``dpi`` and ``language`` are passed thru from the respective function parameters.
73+
74+
75+
Selecting Pages for OCR
76+
------------------------
77+
78+
Usually in document processing, the vast majority of pages contain extractable text and do not require OCR. PyMuPDF4LLM contains logic that analyzes the content based on a number of criteria including (but not restricted to) the following:
79+
80+
* Presence of extractable and legible (!) text
81+
* Presence of images that appear to contain text
82+
* Presence of vector graphics that simulate text
83+
* Presence of text generated by previous OCR activities
84+
85+
The OCR decision is internally based on the results of the following function::
86+
87+
from pymupdf4llm.helpers.utils import analyze_page
88+
89+
analysis = analyze_page(page)
90+
91+
The result ``analysis`` is a dictionary with the following keys and values. The area-related float values are computed as fractions of the total covered area.
92+
93+
* "covered": pymupdf.Rect, page area covered by content
94+
* "img_joins": float, fraction of area of the joined images
95+
* "img_area": float, fraction of **sum** of image area sizes
96+
* "txt_joins": float, fraction of area of the joined text spans
97+
* "txt_area": float, fraction of **sum** of text span bbox area sizes
98+
* "vec_joins": float, fraction of area of the joined vector characters
99+
* "vec_area": float, fraction of **sum** of vector character area sizes
100+
* "chars_total": int, count of visible characters
101+
* "chars_bad": int, count of Replacement Unicode characters
102+
* "ocr_spans": int, count: text spans with ignored text (render mode 3)
103+
* "img_var": float, area-weighted image variance
104+
* "img_edges": float, area-weighted image edge energy
105+
* "vec_suspicious": int, minimum number of suspected vector-based glyphs
106+
* "reason": str, reason for the OCR decision, else ``None``
107+
* "needs_ocr": bool, OCR decision (recommendation)
108+
109+
The reason is one of the following values:
110+
* "chars_bad": more than 10% of all characters are illegible (i.e. Replacement Unicode characters)
111+
* "ocr_spans": there exist text spans created from previous OCR executions (render mode 3)
112+
* "vec_text": there exist suspected vector-based glyphs
113+
* "img_text": there exist images which probably contains recognizable text
114+
115+
Based on this analysis, PyMuPDF4LLM will decide whether to invoke or skip OCR for a page. This is done to optimize processing time and resource usage by only performing OCR when it is likely to yield additional text content that cannot be extracted by other means.
116+
117+
You can override this logic in the following ways:
118+
119+
1. By setting `force_ocr=True` in the output functions (`to_markdown`, `to_text`, `to_json`). All pages will then be OCRed with the selected or provided OCR function regardless of their content. This will obviously have a massive impact on your execution time: expect several seconds duration per each page.
120+
121+
2. Do as before, but add your own selection logic to the OCR plugin::
122+
123+
import pymupdf4llm
124+
from pymupdf4llm.ocr import rapidocr_api
125+
from pymupdf4llm.helpers.utils import analyze_page
126+
127+
def my_ocr_function(page, pixmap=None, dpi=300, language="eng"):
128+
# analyze the page content and perform OCR only if necessary
129+
analysis = analyze_page(page)
130+
if not analysis["needs_ocr"]:
131+
return None
132+
133+
# if OCR is recommended, you can decide differently based on your own insights, e.g.
134+
if analysis["reason"] == "ocr_spans":
135+
# we might want to accept previous OCR:
136+
return None
137+
138+
rapidocr_api.exec_ocr(page, pixmap=pixmap, dpi=dpi, language=language)
139+
return None
140+
141+
md_text = pymupdf4llm.to_markdown("input.pdf", force_ocr=True, ocr_function=my_ocr_function, ...)
142+

0 commit comments

Comments
 (0)