How to use Umi-OCR: install, screenshot OCR, batch and PDF tips

Get Umi-OCR and start it

The README describes Umi-OCR as a download-app: you get a release package rather than running an install command. Release packages are provided as a .7z archive or a .7z.exe self-extracting file. The self-extracting version can be unpacked on a computer that does not have an archive tool installed.

The program does not need to be installed. After extracting, double-click Umi-OCR.exe to start it. The README lists Windows 7 x64 and Linux x64 as supported platforms.

• Windows: extract the package, then run Umi-OCR.exe.
• Scoop users (Windows): the README lists two options, extras/umi-ocr (bundled Rapid-OCR engine) and extras/umi-ocr-paddle (bundled Paddle-OCR engine). Do not install both, because shortcuts may be overwritten.
• Linux: the README's file tree lists umi-ocr.sh; build and runtime setup for Linux is covered in the linked runtime repository.

Recognise text from a screenshot

The Screenshot OCR tab lets you press a shortcut to capture part of the screen and recognise the text in it. The left preview panel lets you select and copy from the image directly. The right result panel lets you edit recognised text and select several records to copy at once.

You can also copy an image from another place and paste it into Umi-OCR to recognise it.

• Open the Screenshot OCR tab and use its shortcut to capture the screen.
• Edit text in the result panel if the recognition needs fixing.
• Paste images copied from elsewhere if you do not want to take a screenshot.

Choose a text layout option

Post-processing controls how recognised lines are arranged. The default is suitable for most cases: multi-column layout detection with line breaks at natural paragraphs. The README also lists options for single-column text, line breaks after every sentence, no line breaks, and keeping indentation for code screenshots.

The README states these options handle horizontal text and vertical text read right to left, though vertical text also needs support from the OCR engine itself.

• 多栏-按自然段换行: suitable for most cases, multi-column layout, natural paragraph breaks.
• 多栏-无换行: merges all sentences into one line.
• 单栏-保留缩进: intended for code screenshots, keeps leading indentation and spaces.
• 不做处理: raw engine output, each sentence on its own line by default.

Batch-process many image files

The Batch OCR tab imports local image files for recognition. The README lists supported input formats as jpg, jpe, jpeg, jfif, png, webp, bmp, tif and tiff. Results can be saved as txt, jsonl, md or csv (Excel). The README says there is no limit on the number of images, and that hundreds can be imported in one task.

For very large or tall images, the README suggests raising the image side length limit under the page settings in the text recognition section. You can also set the computer to shut down or sleep when a task finishes.

• Input formats: jpg, jpe, jpeg, jfif, png, webp, bmp, tif, tiff.
• Output formats: txt, jsonl, md, csv.
• No count limit stated; the README mentions hundreds of images in one task.
• Adjust 限制图像边长 (image side length limit) if large images fail.

Remove watermarks with ignore regions

Ignore regions are a batch OCR feature for excluding unwanted text. Open the ignore region editor from the right-hand settings of the Batch OCR tab. Hold the right mouse button to draw one or more rectangles. Text inside these areas is skipped during the task.

The README warns that a whole text block must sit inside the rectangle to be ignored; single characters are not filtered individually. Draw boxes generously so they cover every place a watermark might appear.

• Open the ignore region editor in the Batch OCR right-hand settings.
• Hold the right mouse button and draw rectangles.
• Make boxes large enough to cover all possible watermark positions.
• Only text blocks fully inside a box are ignored.

Extract text from PDFs and scanned documents

The Document OCR tab accepts pdf, xps, epub, mobi, fb2 and cbz files. It can run OCR on scanned pages or extract existing text. The output can be a double-layer searchable PDF. Ignore regions can also be set here to exclude header and footer text, and the computer can be set to shut down or hibernate after the task.

• Input formats: pdf, xps, epub, mobi, fb2, cbz.
• Output: text, or a double-layer searchable PDF.
• Use ignore regions to exclude headers and footers.

Read and create QR and barcodes

The QR tab has two modes. Scan mode reads QR codes and barcodes from a screenshot, pasted image or dragged-in file, and supports more than one code per image. Generate mode turns text into a QR code image, with options such as error correction level.

The README lists 19 supported formats, including QRCode, DataMatrix, PDF417, Code128 and EAN13.

• Scan: screenshot, paste or drag in an image.
• Generate: enter text and choose the error correction level.