Persian image and PDF to text, with tables and charts

Give it one page and get its text back in reading order. A table stays a table, a chart’s numbers become a table and are drawn again, and a flowchart is written step by step with its arrows.

  • Free, up to 6 pages a day
  • Photo, scan or PDF
  • Files are not kept

The chart becomes a table of numbers and is drawn again; a value not printed on the chart is marked ≈.

Reading

[نمونه]

شرکت نمونهٔ آریا پخش — واحد برنامه‌ریزی و تحلیل فروش

گزارش عملکرد فروش نیمهٔ نخست ۱۴۰۴

شمارهٔ گزارش: ۱۴۰۴/ف/۰۷ — تاریخ تهیه: ۱۵ مهر ۱۴۰۴

۱- خلاصهٔ عملکرد

در شش ماه نخست سال ۱۴۰۴ تعداد سفارش‌های ثبت‌شده به ۵۸٬۴۲۰ سفارش رسید که نسبت به دورهٔ مشابه سال قبل ۱۸ درصد رشد نشان می‌دهد. بیشترین رشد در ماه‌های خرداد و تیر و همزمان با آغاز طرح فروش اقساطی رخ داد. افت مرداد عمدتاً ناشی از تعطیلی‌های رسمی و کاهش ساعت کار مراکز توزیع بوده است.

نمودار ۱- تعداد سفارش‌های ماهانه (هزار سفارش)

ماه۱۴۰۴ (هزار سفارش)۱۴۰۳ (هزار سفارش)
فروردین۸٫۱≈۶٫۸
اردیبهشت۸٫۹≈۷٫۵
خرداد۱۱٫۲≈۸٫۱
تیر۱۱٫۹≈۸٫۸
مرداد۸٫۷≈۸٫۴
شهریور۹٫۶≈۹٫۶
محور افقی
؜ماه
محور عمودی
؜هزار سفارش
منبع
؜سامانهٔ ثبت سفارش شرکت؛ ارقام ۱۴۰۴ قطعی و ارقام ۱۴۰۳ از گزارش سالانه.

خلاصهٔ نمودار: بیشترین تعداد سفارش ماهانهٔ ۱۴۰۴ در تیر با ۱۱٫۹ هزار و کمترین در فروردین با ۸٫۱ هزار بوده است.

۲- سفارش‌ها به تفکیک منطقه

منطقهتعداد سفارشسهم از کل (درصد)تغییر نسبت به ۱۴۰۳
تهران و البرز۲۴٬۱۸۰۴۱٫۴‎+۲۲٪
شمال‌غرب۹٬۷۶۰۱۶٫۷‎+۱۴٪
شرق۸٬۹۳۰۱۵٫۳‎+۹٪
جنوب۸٬۲۵۰۱۴٫۱‎+۲۵٪
غرب۷٬۳۰۰۱۲٫۵‎-۳٪
جمع۵۸٬۴۲۰۱۰۰‎+۱۸٪

گزارش داخلی — فقط برای استفادهٔ مدیران

صفحهٔ ۳

Sample sales report page with a two-series column chart and a table of orders by region

An invented page we made for this sample; the reading beside it is exactly what the reader wrote, unedited.

What it reads

01

Persian and English text

In reading order, with half-spaces, Persian dates and digits kept as printed.

02

Tables

Every row and every cell, empty ones too. A table stays a table, and its columns do not run together.

03

Charts

The chart’s form and the value of every bar or slice. A value not printed on the chart is read off the axis and marked ≈.

04

Flowcharts and diagrams

Steps with their arrows, yes/no branches, loops and phases. An organisation chart is written as a nested list.

05

Stamps, signatures and logos

Each is marked where it sits, as [مهر] or [امضا].

06

Unreadable words

Not guessed: written as [ناخوانا], so you know where to check against the original.

How this differs from plain OCR

Plain OCR pulls letters out of a picture. This reader looks at the page the way a person does: it sees where the table is, where the chart is, and which arrow goes to which box.

Where it can be wrong

A value read off a chart’s axis is approximate, which is why it carries ≈. Handwriting and poor scans have more errors than print. For an important document, check the text against the page; that is why the page is shown beside it.

Plain OCRReading with a model
Returns the text line by line; table columns run together.Writes a table as a table, with every row and column.
Many Persian PDFs store their text layer reversed or letter by letter.Reads the page from its picture, not the file’s text layer.
Skips charts and diagrams, or keeps only their labels.Writes a chart’s numbers and a flowchart’s structure.
Replaces a word it cannot see with wrong letters.Marks an unreadable word and an uncertain arrow.

The same reading, for all of an organisation’s documents

This page reads one page at a time. In Zharfban, the same reader goes through a whole document page by page, and you can ask questions of all of it; answers come with the page number and the document’s own words.

  • Whole documents read page by page in the background
  • Questions answered with the page and an exact quote
  • Installed on your own servers, so documents stay inside
Request a discovery callDocument automation and Persian OCR for organisations

Persian OCR: frequently asked questions

Is it free?

Yes. Each internet address can read up to 6 pages a day and 3 an hour, one page at a time. For whole documents or volume, contact us.

Does it read Persian handwriting?

It usually reads legible handwriting, with more errors than print. A word it cannot read is not guessed; it is written as [ناخوانا].

How do I read a multi-page PDF?

Choose the PDF and pick the page. The page is turned into a picture in your browser, and only that page is sent to be read.

Where does my file go, and is it kept?

The page is scaled down in your browser and sent only for this reading to an AI service outside Iran. We do not store it, and under that service’s commercial terms the data is not used to train models. Do not send confidential documents here; for those we install the reader on your own servers.

How accurate is it?

There is no fixed figure: it depends on the image quality, the font or handwriting and the layout. That is why the text sits beside the page, every approximate value carries ≈ and every unreadable word [ناخوانا].

What about volume, or connecting it to our systems?

The same reading can run over your documents and deliver its output to your accounting system, archive or work queue. A discovery call settles the document types, volume and where it runs.