How to use the image text extractor
- Pick the language to recognize. For images that mix Korean and English, the default Korean + English works well.
- Choose or drop an image. You can also paste a screenshot. The first time takes a moment because the files needed for recognition have to download.
- Check the text below, fix any mistakes, then copy it or save it as a txt file. If the result looks odd, change an option and press Recognize again.
How text in a photo is read
OCR is the technology that turns the shapes of letters in a picture into text a computer can handle. This tool first finds the areas and lines that contain text, cuts the picture line by line, and estimates which characters follow one another in each line. Because it looks at the flow of the whole line instead of guessing one character at a time, it can tell similar-looking letters apart by their context.
Before recognition the tool applies three adjustments. Small images are enlarged to a comfortable reading size, color is converted to grayscale with more contrast, and mostly dark images are inverted so the text is dark on a light background, which reads best. Below the result a recognition confidence value is shown. Under 60% usually means the text is blurry or too small, so it is better to try again with a sharper image.
All of this is calculated inside your own computer or phone. The picture is not sent to a server, so there is no upload wait and the photo never leaves your device. The first time you need to download the files needed for recognition and the language data, and afterwards they are kept by your browser so later starts are faster.
Ideas and limits
You can photograph a paper document and copy sentences, or save the effort of retyping text from a screenshot into a notes app. Results are best on images where printed text is clear, such as receipts, notices or a paragraph from a book. The recognized text can be edited right in the text box, which is much faster than typing a document from scratch.
When taking the photo, shoot straight on in a bright place so the document fills the frame and no shadow falls on the text. Text that is tilted or curved because of an angled shot lowers recognition a lot. Straightening the picture in a scanner or photo editing app before uploading can improve results noticeably.
It is built to read printed text, so handwriting, cursive and highly decorative fonts are less accurate. It does not understand table cells and reads from top to bottom, so it is not suited to copying a table as it is. Korean has many characters with similar shapes and final consonants, so misreads happen easily. Always compare important information such as numbers, accounts and contact details with the original.
Sometimes unnecessary spaces appear between letters or line breaks cut a sentence in the middle. Turning on Tidy line breaks replaces line breaks inside a paragraph with spaces so it flows as one paragraph. Blank lines between paragraphs remain. For poems and lists where line breaks carry meaning, leave this option off.
Very large images take a long time, so pictures whose long side is over 3,000 pixels are reduced before reading. Images whose long side is under 1,600 pixels are enlarged up to three times. A photo taken with a phone is usually fine as it is, and very small screenshots do better with enhancement on.
Processing happens only on your own device, which is an advantage for images with personal information, but after you copy the result or save it as txt you are responsible for how that file is stored and shared. If it contains sensitive information such as ID or account numbers, double check where you share it.
Tips for better recognition
- Choose a sharp, large image. Enlarging a blurry photo does not help.
- Shoot straight on without tilt and avoid shadows.
- Pick Korean + English for mixed text, or just the one language if there is only one.
- If the result looks odd, toggle image enhancement and recognize again.
- Compare important details such as numbers and contacts with the original once more.
FAQ
Is my photo uploaded to a server?
No. The files needed for recognition and the Korean and English language data are served as files of this site and all calculation happens inside your browser. The photo is not uploaded anywhere, so even photos of contracts or receipts can be handled with more peace of mind.
Why does it take longer the first time?
The first time, the files needed for recognition (about 4 MB) and the language data (about 1.5 MB for Korean and 3 MB for English) have to be downloaded. After that your browser keeps them, so later starts are much faster.
Korean text comes out wrong or gets extra spaces between letters.
Recognition depends heavily on how sharp the picture is. Small or blurry text can be read as similar-looking characters or get unnecessary spaces. Use a large image close to the original when you can, and correct the result in the text box.
Does it recognize handwriting or tables?
It is built to read printed text, so handwriting is much less accurate. It does not understand table cells and reads text from top to bottom in order, so it is not suited to copying a table as it is.
What about light text on a dark background?
With image enhancement on, a mostly dark image is inverted automatically before reading. If the result still looks odd, try again without enhancement, or raise the contrast between text and background in a photo editor first.
Notes & limitations
This is a simple free text recognition tool and errors can occur depending on the photo. When you use the result for important decisions such as contracts, finance or legal matters, check it against the original. Photos are processed only in the browser and are not sent to a server. Check the laws and rights before using images that contain other people's works or personal information.