ChatGPT must improve the OCR scanning method

ChatGPT Feature requests
1

The problem is that ChatGPT fails to scan the files properly.

I have attached 2 identical pages from a pdf with old pages. The first page is the original one from the PDF (you can see the writing a bit poorly), the other is the same page after it was saved in TIFF, from the pdf. You will see the difference immediately.

Let’s say that I have a PDF file with such images that are not clear, or because the pages of the book sound old, damaged or printing problems. In order to visualize them much better, because many of the books have writing barely visible, i.e. erased, then you have to save each page from the pdf in the .tiff extension

This will make the writing darker and much clearer. Only then go through OCR the image. Adobe Acrobat Pro - Save as other - Image - TIFF All pdf pages will be automatically saved in TIFF, and each page will be seen 68% clearer. Any AI tool must do the same before scanning with OCR, because it will be much easier to read the writing.

As a basic idea, if you want to do a quality job, for example if you are crawling with OCR among thousands of PDF files, so in order to correctly process a text from a pdf (with OCR), you have to scan it several times times, like this: first with thin text, then with thick text, then without shadows, then with more brightness.

Original image, bad quality

versiune 1
versiune 1590×846 97.2 KB

Same image, good quality:

versiune 2
versiune 2509×770 80.1 KB

Test with ChatGPT, see if it can correctly identify the text from the original page (with bad quality), then how to identify the text from the same pdf page, but saved in .tiff format

2

“Improving your OCR ability for difficult-to-read Romanian texts”
Objective: To improve your ability to transcribe texts from blurred or damaged images, with an emphasis on the Romanian language and historical documents.
Process:

You will receive an image containing text in Romanian.
You will analyze the image and transcribe the text as accurately as possible.
I will give you corrections and clarifications for the words you didn’t identify correctly.
You will add diacritics and correct the transcription according to my feedback.
You will use this information to “recalibrate” your recognition ability, creating an improved template for future analysis.

Benefits:

You will improve your accuracy in recognizing old or unclear Romanian texts.
You’ll create a reference database for hard-to-recognize words and phrases.
You will develop your ability to contextually and linguistically interpret partially illegible texts.

Through this iterative process, you will learn to more effectively identify the specific characteristics of Romanian texts in various stages of preservation, thus improving your transcription accuracy for future similar tasks.

For example:

image

ChatGPT only managed to identify 90% of the words. The exact text in the images is as follows:

bine și se acordă cu formula faimoasă care i-a servit mult timp de deviză: „Laisser faire, laisser passer”.

Partizanii ei se apără cu tărie și pretind să reprezinte însăși știința. Își dau chiar ei titlul scurt de tot, „economiști”, și adversarii lor le admit aceasta în cele mai multe cazuri. E adevărat că originile acestei școli se confundă cu acele ale științei economice înseși. Doctrina ei e foarte simplă și se poate rezuma în trei puncte:

  1. Școala liberală crede în existența unei ordini naturale, în sensul că societățile umane sunt guvernate de legi naturale pe care nu le putem schimba când vrem noi, pentru că nu le-am făcut noi și, de altminteri, n-avem interes să le modificăm, chiar dacă am putea, pentru că sunt bune sau, cel puțin, cele mai bune posibile. Sarcina economistului se mărginește la descoperirea jocului acestor legi naturale și datoria indivizilor și a guvernelor e să-și dea silința să potrivească atitudinea după dânsele.

As I said, in order to improve ChatGPT ability to read with OCR, it must form recognition templates. I give the exact words from the pictures, and in the future ChatGPT will have to know how to recognize the letters and words from other pictures that are not very clearly visible. ChatGPT will compare the words that I see well, with the words that he did not identify well, and thus a template is formed.

3

Therefore, to recognize as many forms of writing as possible, someone needs to send you the image with the text, then send you the legible text, which you must superimpose over the text in the image, and outline it, placing and positioning letter over letter, word over word, so that you manage to form the pattern of each letter.