CodeKitHubHow to Use
- Take a photo of the book page — try to fit the whole page in frame, but don't worry about holding the camera perfectly level.
- Click or drag the photo into the upload area.
- Leave "Auto-correct skew" checked (it's on by default) and pick the page's language.
- Click "Extract text" — the tool straightens the image first, then recognizes the text.
- Copy the result or download it as a .txt file.
Example
Input
A phone photo of an open paperback page, held at roughly a 10° angleOutput
Detected skew: -10.5° → text extracted with full paragraph structure preservedWithout the deskew step, a photo at this angle typically loses a third of its words to misreads; correcting the angle first brings recognition back to near-scanner accuracy.
What Is This Tool?
This tool extracts text from a photograph of a book, textbook, or printed page using OCR (optical character recognition) that runs entirely inside your browser. Unlike scanner-based OCR tools, it's built around the specific problem of hand-held photos: pages photographed at a slight angle, which is nearly unavoidable when you're holding a phone over an open book instead of pressing it flat onto a scanner glass.
Before recognition even starts, the tool analyzes the image to detect exactly how many degrees it's rotated from horizontal, then straightens it automatically. This single step matters more than almost anything else for photo-based OCR: a page photographed at just 10-12 degrees off-level can drop recognition accuracy from over 95% to below 70%, turning clearly readable text into garbled nonsense — deskewing first restores it to near-scanner-quality results.
Why photo angle matters more than you'd think
OCR engines read text by scanning pixel rows looking for consistent horizontal bands of ink — that's how they separate one line of text from the next. When a page is tilted even slightly, what should be a clean horizontal line of text becomes a diagonal smear across many pixel rows, and the engine's line-segmentation logic starts merging characters from adjacent lines or splitting single characters apart.
This is why a photo that looks perfectly readable to your eyes can still produce garbled OCR output: your brain effortlessly compensates for a slight tilt, but a naive OCR pipeline does not. The fix isn't a smarter recognition model — it's straightening the image first, which is exactly what the auto-correct step in this tool does before recognition ever runs.
Common use cases
- Digitizing notes or excerpts from a physical textbook without retyping them by hand.
- Extracting a passage from an old paperback or library book to quote or search later.
- Converting a photographed handout, worksheet, or printed article into searchable, editable text.
- Archiving pages from a book that's falling apart or too fragile to scan flat on a scanner bed.
Getting the best results
- Good, even lighting matters more than a perfectly steady hand — a slightly crooked but well-lit photo out-performs a level photo with heavy shadow across the text.
- Avoid direct flash on glossy pages; the resulting glare/hotspot can blank out the text underneath it regardless of angle correction.
- Fit the whole page in frame with a bit of margin — cropping too tightly can cut off the first or last line.
- For very long passages, photograph and process one page at a time rather than trying to capture a two-page spread, which introduces its own curvature distortion near the spine.
Why Use It?
Automatic skew correction: detects and straightens crooked photo angles before recognition, the single biggest accuracy factor for hand-held book photos.
Fully offline processing: the photo and extracted text never leave your browser — nothing is uploaded to a server.
Works for old, yellowed, or slightly faded book pages as well as crisp modern print.
Supports multiple recognition languages (English, Chinese, Spanish, French, German, Japanese) so you can scan books in different languages.
No account, no upload limit, no subscription — free for as many pages as you need.
Frequently Asked Questions
Why does my book photo need special handling instead of regular OCR?
Regular OCR tools are usually tested and tuned on flat, perfectly aligned scans. A book page photographed by hand is almost never level — even a small 8-12 degree tilt, which is normal for a handheld shot, can crash recognition accuracy well below 70%. This tool measures that tilt and corrects it automatically before running recognition, which is the difference between usable text and garbled output.
Does this work on old or yellowed book pages?
Yes — page discoloration and mild contrast loss from aging paper have very little effect on recognition accuracy by themselves. The angle of the photo is the dominant factor, not the paper's color, so old paperbacks and used textbooks work about as well as new ones once the image is straightened.
Is my photo uploaded anywhere?
No. Both the skew detection and the text recognition run entirely in your browser using WebAssembly — nothing about your photo or its content is sent to any server. You can even disconnect from the internet after the page first loads and it will keep working (aside from the one-time download of the recognition engine).
Can I turn off auto-correction if my photo is already straight?
Yes, untick "Auto-correct skew" if you're working from a flat scan or a photo you've already straightened — this skips the detection step and runs recognition slightly faster.
Why is the first extraction slow?
The first time you click "Extract text", the tool downloads the OCR recognition engine (a few megabytes, one time only, cached by your browser afterward). Every extraction after that is much faster since the engine is already loaded.
What languages can it recognize?
English, Simplified Chinese, Spanish, French, German, and Japanese are available in the language selector. Pick the language the page is actually printed in for the best accuracy — recognition quality drops sharply if the wrong language is selected.