How can a scanned copy be turned into editable text? After typing two pages of the test paper, I remembered that WeChat was hiding free OCR

the second week of school, the homeroom teacher posted in the group chat, asking parents to organize their child's mistakes from last semester into an electronic version for the practice test. I patted my chest and took the job, opening Word into the four test papers I had taken on my phone, typing each word into the paper. After finishing two word problems, 40 minutes were gone, and my neck was stiff. The kid leaned in to look and said, "Dad, can't you just hold down this character and extract it?"

Alright, I got a lesson from an elementary school student. I'm not very experienced at scanning. The ink warehouse all-in-one at home usually scans documents and medical reports. I wrote a complete guide on scanning before, how to turn paper into images. But after the paper becomes an image, I never seriously figured out how the text in the image becomes editable text. That night, I ran through all three free routes, and for the second round, I had six papers left, ten minutes in total.

first check what type of document you have, don't recognize it right away

can't skip this step. The first time, I didn't distinguish the file type and wasted effort. There are basically three types of things in his hand.

The first type is photos and screenshots—JPG or PNG, one by one. There's no ambiguity here; to cut out the text, you have to rely on recognition. The second is PDF, but PDF needs to be split into two parts: drag the text with the mouse, and you can pull up the selection and copy it—this is the text version. The text version is originally arranged from the text, and just press Ctrl+C to get it—there's no need for recognition to appear. If you can't drag out the selection or just click to treat the whole page as a single image, that's a scanned version. Essentially, the image is wrapped in a PDF shell, treating it the same way as the photo. The third method is the original paper version, which requires taking photos or scanning to create an electronic image. This step can be done with a printer or phone.

judgment is simple: what you can select isn't recognition, it's copying. This tool only serves images and PDF-formatted images.

First Way: Long-press 'Extract Text' in WeChat, the ceiling for emergencies

this road doesn't need to be packed with anything. In WeChat, open that image—send the photo to the file transfer assistant, or you can dig through the original image in chat history—enlarge the image, long press with your finger, and the menu will show the words "Extract Text." Results in one second—select, copy, and share the full text freely, unlimited times, not even internet restrictions.

I tested my child's test paper: a well-photographed, well-lit test with about 600 words. After recognizing and reading through once, I made 4 mistakes. The mistakes were all familiar faces—the number 1 was mistaken for lowercase l, 0 for uppercase O, a comma was missing, and another '己' was mistaken for '已'. The pattern is clear: the main Chinese text is basically all correct, but all the mistakes come from numbers, letters, and punctuation. So after recognition, don't just submit errors; spend a minute reading through, focusing your energy on numbers and punctuation, especially the numbers you need to fill in the table.

shortcomings are also obvious. You can only handle one card at a time. When you have over twenty cards, you can open it, long-press, copy, cut windows, paste it—one whole set of clicking can make anyone's hands cramp. Tables are its weak spot. The 8-row, 6-column course schedule on the test papers can be recognized as collapsing into rows of text, with no vertical lines, so you have to re-engineer which columns to match.

positioning clearly: one or two images to get the chance, WeChat is so convenient that there's no competitor. Once the volume is too large, it can't hold up.

the second path: Lots of forms and layouts, special tools like line tracing are put into play

put a 'Baimiao' on your phone—that's for reading tables. Same class schedule: long-press in the album to enter batch mode, then after recognition, click "Table Recognition" to restore it to an exportable table: 8 rows and 6 columns arranged horizontally and vertically. I barely edited it into Excel. When taking pictures of it, make sure the paper is straight and all four edges are included in the frame, and the success rate of reproduction can be even higher.

The

price is the free quota: five recognition attempts per day, one batch call, and members can unlock it for just over ten yuan. For someone like me who focuses on work once a semester, the free quota is actually enough; In office scenarios where you have to scan contracts every day, buying a membership is easier than anything else. Also, remember one detail: it uses cloud recognition, and the image must first be sent to the server's server for calculation before returning—fast as it is, but keep this detail in mind for now; we'll test it in the next section.

to add, many phone albums now have this feature too. On Android, there is a "T" recognition entry in the image interface—click the frame to copy it, and the effect is on par with WeChat. If you have something on hand, just use it directly—no need to judge by superiority.

the third option: Sensitive documents and contracts, I only use the offline Umi-OCR

third path is the last one I stayed to be the main force. Umi-OCR is open source, free, offline, runs on Windows, and its core is PaddleOCR. These words each describe their weight: Open source and free means no membership, no usage limits, no watermarks; Offline means you can't get the computer—ID card, household registration book, purchase contract, bank statements, and so on. In principle, I don't let them go to any cloud platforms, including the handy app from the previous section.

usage is almost excessively simple. After installation, there's a small window—drag the image in and the text will appear; Not afraid of a whole bundle, I picked them all and dragged them in together, and it lined up and ran on its own. I scanned my 32-page manual, dragged it in, soaked it in a cup of water, and brought it back. Everything was done, and it was output as text that could be pasted directly.

another high-frequency use I didn't think of is "screenshot OCR": set a shortcut key, and anywhere on the screen—text on images or videos that are not allowed to be copied, subtitles in videos, pop-up error boxes—press the shortcut to lock the frame, and the text goes straight to the clipboard. It's two steps less than taking a photo and then sending it to your phone. Now, I rely on it to check error messages. You can drag the entire scanned PDF into it, and arrange it as text for you. Batch recognition takes a lot of CPU time. My old desktop in my study takes two or three seconds per page—not exactly fast, but it doesn't require anyone to watch over it. You can drag it in before bed and receive it in the morning.

handwriting and math formulas, lower expectations before taking action

pick any of the three paths, and if there are two types of things, you need to get a heads-up first.

handwritten style is the first pitfall. I tried to identify a passage of handwritten comments from the teacher's error notebook, and half of the answers depended on the context to guess. The phrase "得地得" was a chain of errors, forming a sentence I didn't even recognize. Grandpa's penmanship was even less promising; the handwriting on the scanned copy faded and was immediately slacking in recognition. Handwritten materials are most efficient when the human eye slowly taps; machines are only suitable for recognizing printed fonts and clear printed text.

mathematical formula is the second pit. The top and bottom structures of the multiplication sign, square sign, and fraction line are either question marks or a string of garbled text. The square of x can be written as x², which looks so obvious but sticks it into Word and is completely exposed. For the formulas in the error book, my only method is: honestly type by hand, or use the input method's built-in formula panel. Competing with machines like this isn't worth it.

by the way, images with vertical fonts, WordArt, or fancy backgrounds will also lose accuracy. I dug out a few old photos, with the back showing the date my grandfather wrote in pencil on the shoot, vertical in portrait rows. I tried three shots but none were ruined. Later, I used an old tablet to make an electronic photo frame hanging in the entryway slowly flipping through it. The text on the back was left for the eyes to recognize; if not, it was just a keepsake.

take the same path as the document is, copy it according to this table

Review the

order: After picking up the item, try using the mouse to select the words. If you can pick one, copy and finish the work; If you can't select it, see what kind of product it is—just long-press one or two images on WeChat to solve it; Upper white drawing with tables; If you come across a document, contract, or a stack of scans, open Umi-OCR and drag it in. After recognizing, I always read through the text, staring at the numbers and punctuation marks—I couldn't skip a minute.

wrapped up and explained the consequences of that error notebook: after identifying, proofreading, and formatting, I saved it as a PDF, connected my phone to the printer typed a copy directly and brought it to school. The mistakes on the paper circled around and then returned to the paper. That loop saved me two nights of typing time.