Home / Docs / Download / Release Note / Blog / ImageTrans / Support / Donation / RSS
ImageTrans - Computer-Aided Image and Comic Translation Tool
ImageTrans is a computer-aided image and comic translation tool. It can automatically locate text areas and perform OCR operations using state-of-art OCR technology and a homebrew text areas merging and detecting algorithm, which is specially designed for comics (also webtoon, manga, manhwa and manhua). The original text can be automatically erased and the translation be reinjected.
ImageTrans can be used not only as an image translator, but also an image reader, an image transcriber, a screenshot OCR and translator, and a deep learning annotator.
ImageTrans is the result of the thesis for my master’s degree: Design and Implementation of a Computer-Aided Comics Translation Tool.
Accurate text area detection
It can use natural scene text detection, object detection and a heuristic method to detect text.
Accurate text recognition
Support multiple OCRs
It supports online OCR services such as Google, Baidu, Sogou and Youdao, and offline OCR engines such as Tesseract, PaddleOCR and Windows 10’s built-in OCR engine. See FAQ for the usage and the list of supported OCRs.
In addition, you can also call other OCR services by writing plug-ins.
Support major languages
The OCRs used support all the major languages in the world.
Employ a variety of image processing methods to improve the results
Strip Japanese furigana
Sogou OCR: えーっとはいばん灰原さんのせき席は…
Sogou OCR: えーっと灰原さんの席は…
Convert vertically aligned images to horizontally aligned
Most of the OCR models based on CRNN can only recognize text horizontally aligned, so it is necessary to convert vertical images to horizontal images when using these OCR models
Baidu OCR: 強んな勉とら!文事なかす原るん日で灰すみ今
Baidu OCR: 今日からみんなと勉強する事になった灰原哀さんです!
- Post-processing of the text: remove line breaks, spaces
Use spellcheck to improve the result
It can call a variety of machine translation engines. It also supports translation memory, corpus concordance, term management and other common functions of computer-aided translation software. It has a sound effects words search engine designed for translating comics.
Precise text removal and reinjection
The software can use tools and algorithms like Sickzil-Machine, binarization, PatchMatch to generate text mask and remove text. It can refill the translation and automatically adjust the font size according to the position and area of text regions. All the above operations support manual adjustment.
- It can use scripts to save the results as Photoshop’s PSD files.
- It can export the data to Excel, Word, XLIFF files, or import data from these files
- With the Chrome Extension, it is possible to translate pictures on web pages directly
Like BasicCAT, ImageTrans is cross-platform (Windows, MacOS, Linux). It can also run as a server for online usage.
- Global font style
- Automatic detection of background color and text color
- Batch processing
Examples of Translated Comics
Silver polish: https://blog.xulihang.me/Comics-Viewer/viewer.html?uniquename=silverpolish
Example 2 (old):
Come to me, Love —— From issue 41, Boy Loves Girls (Golden age comics).
Browsing Page: http://comics.xulihang.me/viewer.html?project=come_to_me_love.itp
Text Searching Page: http://comics.xulihang.me/search.html
Get a Copy
ImageTrans is available for purchase on FastSpring: link.
- ImageTrans for personal use: $10.99
- ImageTrans for commercial purpose: $250
Free update within half a year.
Please follow the installation instruction.
If you are from academic institutions, you can email to firstname.lastname@example.org with your institution’s email to get a discount. Please tell us who you are and what you are gonna do with ImageTrans.
If you cannot complete the payment, please contact me.
- Update the ChatGPT plugin to support batch translation by images #issue404
- The target text can be empty when importing from TXT files
- If an error occurs importing from TXT files, prompt which line is wrong #issue412
- New mask generator plugin: InsetRect #issue413
- Fix a bug that the text is cleared after resetting the target position with the autosaving of source text and target text enabled #issue408
- Add a context menu item to delete terms
- Fix a multi-line matching problem for Find and Replace’s regular expression mode #issue402
- New export option: TMX (translation memory exchange file)
- Update the ChatGPT plugin to support third-party services #issue401
- Pretranslation can use translation memory for exact matching
- New context menu item: Clear target position #issue393
- Apply the theme for all forms
- Usability optimization of font settings
- Fix a problem that faux italic does not have a correct height for perspective transform when line space is bigger than 0
- It is now possible to adjust the stroke width for faux bold and the obliquity for faux italic
- New export option: PDF #issue366
- New machine translation plugins: ChatGPT, Yandex
- Add rich text editor #issue390
- Add support for faux bold, faux italic, font name and font size for rich text #issue389
- New setting for vertical text engine: characters replacing #issue388
- When multiple textareas are selected, some operations of one textarea are skipped.
- Fix a problem that the target text is not joined after merging text areas
- Add a project setting to generate the text mask and text-removed images by text areas
- Save scene text detection’s text mask if provided #issue370
- Fix a bug that the BBCode is not ignored when measuring the height of the text
- Fix a bug that the BBCode is not effective when the text is set to capital
- Change the default JRE to Liberica JRE 11
- Rich text support (based on BBCode) #issue194
- Fix issues related to network requests, pressing Control to select multiple text boxes and text area’s confidence
- New OCR plugin：macOS’s built-in OCR #issue341
- Support pasting an image with its path
fileto detect whether an image’s format is WebP #issue338
- Fix a bug of pasting an image with data URL
- Fix a bug that the grayscale text mask is passed instead of the original text mask when using third-party inpainters as the default
- Add an option to disable AutoCorrect when inputting
- The shortcut keys can be customized #issue323
- UI improvements
- New feature: set the font style based on the text color detected #issue322
- ScreenReader: Display the screenshot in 100% ratio
- In typesetting mode, hide the borders of text boxes when they are being moved (You can disable this feature in Preferences) #issue314
- New project setting: read subfolders (enabled by default) #issue304
- Fix a problem that if the channel number of a png image is not 4, the inpainting will fail
- Fix the logic of the overlay mask checkbox
- Add an option to export to PNG #issue295
- If the image’s format is PNG, use PNG for the text-removed image and the output image. JPEG will be used for other formats #issue295
- Add a project setting to set the default text box size #issue290
- It is possible to adjust multiple textboxes’ positions using shortcuts #issue285
Fix a data exporting bug which was introduced in v1.7.8 #issue279
- Shortcuts on macOS use the Command key instead of the Control key by default #issue277
- Improve the usablity of the mask editor #issue261
- Add vertical text engine customization settings in project settings #issue271
Improve the behavior of the vertical text engine using different fonts #issue271
- Fix a problem that if the image’s width is too large, the mask editor fails to save the mask image #issue266
- Improve the settings of font styles #issue256
- Improve the drawing mode of mask editor #issue261
- New font style option: kerning (only valid for vertical text) #issue224
- Use perspective transformation to get the corrected image of a rotated text area.
- Other bugfixes #issue246 #issue248
The OCR engine and language used last time will be set as the project’s default. #issue245
Line spacing setting is enabled for horizontal text #issue244
- New text direction option: align text vertically from left to right #issue242
- New preference setting: save OCR results to clipboard #issue169
- It is now possible to adjust the radius of the mask editor’s pen mode
- Fix a bug that when there is no panels detected, exporting to a web page will fail
- Update the JS Photoshop script file to support PSB files
New preferences setting: temp folder for HTTP. It is to solve the “Access is Denied” problem some users encountered #issue218
- Font settings’ UX optimization #issue212
- Shortcut’s optimization #issue205
- Add refresh to the context menu. It can help solve the problem caused by switching images too fast under translation mode #issue211
- New image inpainting plugin：Lama Inpainting #issue216
- Add a manager for original images without text and pure-text images #issue199
- AutoCorrect can be enabled for OCR #issue199
- Fix a problem of importing XLIFF #issue207
- Translation can be imported from a tab-delimitted txt file #issue207
- The background of a font style can be set to transparent #issue208
- Other bugfixes and improvements #issue203
- For cases like translating Japanese manga to English, where the original text is vertically arranged while the target text is horizontal, add an option to adjust the text area’s position and size to fit horizontal display #issue190
- Add Speak buttons under text editing areas (can be hidden) #issue191
- Add options to set the default mask generator and image inpainter #issue192
- Call mask generator and image inpainter to process text areas instead of the whole image to improve speed
- Other minor improvements
- Fix the “open project folder” action and the excessively large rotation mouse cursor for macOS
- Font style selection can be applied to multiple selected text areas and make effect immediately #issue185
- Optimize the display of characters and punctuations for the vertical text engine #issue186
- Fix a vertical text engine’s problem that the text is cut if the font size is large
- Add a new payment platform: Mianbaoduo
- Add free DeepL (https://github.com/xulihang/ImageTrans_plugins/tree/master/deeplfreeMT)
- Webpage export. The webpage supports panel mode and text to speech
- Server can now be used as an HTTP server to host the exported webpage in a LAN
- Vertical text engine: supports line space setting, improves the speed and display of punctuations
- Add full-page text boxes creation
- The output folder’s name can be now customized #issue170
- A completion sound effect can be played after the custom workflow is done #issue171
- The font size in the tool bar can be input directly #issue176
- Text areas can now be rotated by dragging the mouse. The image of the text area will be rotated before OCR to improve the recognition rate #issue157
- Add progress info to batch panel detection and color detection #issue153
- Add the rate setting to TTS. It can read both source text and target text #issue152
- Fix a text index problem for the AU3 Photoshop script #issue160
- Add panel detection for comics. It can be used for sorting text areas and reading on a mobile device by panels #issue147
- Add a context menu item to show the machine translation of the combined text of selected text areas. It is mainly designed to translate sentences splitted into different text areas #issue118
- Other performance improvements and bug fixes
- Balloon detection supports comics which has a high height/width ratio (e.g. webtoon) #issue138
- Auto Font Size supports setting a maximum size #issue146
- ScreenReader can now monitor the clipboard #issue145
- Silent translator: the file path of the source image file will be saved #issue144
- New machine translation plugin: DeepL #issue15
- The expanding operation will check the image size to avoid expanding too much
- Move the scroll position to top after switching image
- Fix a localization problem of the Preferences form #issue141
- Add an option to resize image before balloon detection (scaled to 1024px by default) #issue137
- Update OpenCV to 4.5.5 to support Scaled-Yolo V4 #issue129
- Support export text areas to images and import their OCR results #issue124
- The pure color background used in the non-precise text removing mode can be set as rounded #issue123
- Fix an inconsistency problem of vertical alignment #issue122
- AutoCorrect can be used for machine translation #issue133
- Other improvements on UI and performance
- Add Text to speech (TTS)
- Delete key can now be used to delete text areas
- Timeout setting for heuristic text detection
- Other bug fixes
- Improve the ScreenReader’s behavior issue 110
- Improve the bahavior of textboxes under type setting mode
- Support manual fonts installing issue 100
- Add a compatibility input helper based on Swing JTextArea to serve as a workaround for inputting Tibetan in JavaFX’s TextArea issue 99
- New project setting: Text Editor’s font
- Other bug fixes
- Unify the behavior of the text box in 100% and other display percentages issue 94
- New OCR plugin: Google Drive OCR issue 91
- Fix a Google OCR not getting text problem
- Use thread to avoid the program not responding problem during heuristic text area detection
- Unify the OCR behavior of the screen reader with the main program
- The default display percentage is changed to 100%
- New Tesseract line mode. ImageTrans will extract text lines and use Tesseract to recognize them directly, which can greatly improve the accuracy. (issue87)
- Add an auto capital option
- Add an auto machine translate options to the screen reader
- Other bug fixes
- Add a silent translation helper
- Add a new furigana stripping method based on projection
- Update OpenCV to 4.5.3. The built-in high-level text detection and recognition are supported with the plugin.
- Add a new Japanese OCR plugin：読取革命(Yomikaku).
- Other bug fixes
- Fix a high resolution target image export problem that the width and height of rotated text areas are not correct
- Fix a vertical text’s problem that after a text area too small is expanded, the position of the inner text is not adjusted
- New menu item: open project folder
- Fix a problem that when the toolbar is switched to font settings, it does not load the font settings of the current text area
- Fix a problem that the vertical center button is not updated when a new text area is selected
- Improve the vertical center effect
- Other improvements
- Fix a problem that the program fails to export a target image which has a high resolution (width or height > 8000 pixels)
- AutoFontSize is available for vertically arranged text areas
- Change the text areas adjusting behavior of vertically arranged text areas to meet the change made in 1.4.1 which aligns the text to the right: move the mouse to the left of the text area to resize and the right to adjust positions
- Other improvements
- Add a vertical align option #issue72
- Add a shortcut to remove selected areas #issue71
- If the text mask is not saved as a file and there is no new, deleted or adjusted text areas, the program will not regenerate text-removed images
- Add Hindi, Vietnamese and Indonesian to the default languages list (In the previous versions, we have to input the language codes manually) #issue61
- Vertically arranged text will start from the right by default
- Other improvements
- Add “Save as” menu item
- Add “Navigation” menu
- Add Translated Images Manager. Images marked as translated will be skipped during batch operation (#issue59)
- The Text Editor is dockable
- Use Spinner to adjust the display percentage
- Close the current project automatically when opening a new project (#issue39)
- Save the previous state of ToolBar #issue47
- Fix an exported data containing no translation problem #issue45
- The two context menu items: Delete and invert confidence, now support handling multiple text areas
- When the Control key is pressed, it is possible to select multiple text areas by clicking them
Clear textmenu item (clear source, target or source+target)
- Other improvements
v1.3.5 Update2 (2021/04/26)
- Display a warning when users import pictures without opening a project
- Add JPEG extension
v1.3.5 Update (2021/04/06)
- Language Pair Selector will not appear opening a previous project
- Use the current image’s name as the Image Combobox’s tooltip
- Improve the batch machine translation function. Baidu now supports batch translation.
- Add an Auto Params option to heuristic text localization (experimental).
- Improve global font styles manipulation. It is possible to move them up and down and import styles from other projects.
- Users will be asked to select language upon project creation
- Unify the font color picker
- Improve the undo manager
- Update ImageTrans_OCR. Added ChineseOCR. It now supports a free combination of different text detection and text recognition methods. The plugin is updated as well.
- Fix an index over items size problem during bulk OCR. Related issue
- Add OCR interval setting
- Scale images too small before OCR
- Localize missed layouts
- New OCR plugin: CRAFT+CRNN. This is based on the new ImageTrans_OCR project. It is planned to be a hub of offline OCR engines which have good customizability, accuracy and speed.
- New OCR plugin: Naver’s Clova OCR.
- Support merging and deleting areas with box selection
- Add an option to auto OCR after box creation
- Add shortcuts to manipulate text areas (move and OCR)
- Offline balloon (bubble) detection which is based on OpenCV’s Dnn. Models trained with DarkNet or TensorFlow Object Detection API are supported. Model weights and model config files need to be put under the root of ImageTrans along with a model.json file (example)
- The text editor can display source text and target text in a vertical mode.
- Custom workflow. Batch operations can be customized
- Support importing PDF. PDF files will be rendered to images. Text will be extracted if possible
- Fix a target geometry not being correctly saved problem
- Fix a file path problem of Photoshop scripts
- MaskEditor can generate textmask using selected color. Mask can now be set to translucent.
- Support batch machine translation with ColorfulClouds which will significant improve machine translation speed
- It is now possible to set machine translation interval to avoid failed requests due to too many requests in a short period of time (QPS set by MT services)
- New OCR plugin: ABBYY Cloud
- Support spellchecking of OCR results using LanguageTool
- Flip boxes before sorting them if the reading order is right to left
- Add inpaint radius setting in TextRemover
- Add Theme. Currently, there are dark and green themes available.
- Use my personal keys by default for machine translation services like Baidu, Niutrans and Tencent and OCR services like OCRSPACE, Azure and Baidu.
- Add Undo Manager
- Use TextFlow as the default Text Renderer. Text will no longer be truncated
- Text size and layout will be the same after percent switching
- Other improvements
- Supports running multiple tesseract instances to increase OCR speed
- Save verification for 7 days
- Sogou OCR supports right-to-left text order
- Add Baidu OCR accurate
- Fix Find and Replace problems caused by incorrect box jumping behavior and UI localization
- Convert WebP format images to JPG ones for images downloaded with links (using OpenCV)
- Support stripping furigana in Japanese manga for better ocr results
- Better conversion of vertically displayed Japanese textareas to horizontally displayed ones (rotation and center placement of characters like —)
- Fontstyle Unifying operation
- Textareas generated by OCR will also be expanded
- Fix a target geometry setting problem
- Mask Editor can use the cursor to draw a circle or clear a circle
- Add Capital Letter font setting
- Add minimum width/height overlapped percent setting for text localization
- Fix an overlapped box not being added problem
- The Photoshop script supports styles like bold, italic, capitalization and rotation.
- Add Chrome Extension support. Visit here to see detailed instructions.
- Mask, text-removed and exported images will no longer be imported
- Fix the incorrect behavior of creating new projects which was caused by the recent projects feature introduced in v1.2.2
- Other improvements
- Add Translate after OCR option
- When viewing translated version in non-precise text-removal mode, the text box will be transparent to reveal the source text
- Batch Translation can call balloon detection and scene text detection
- Silent Translator supports font setting
- Add a context menu item to download images from links and add them to the project
- Add Silent Translator to support bulk translation, which can be called through command line or run as a server
- New OCR plugin: sogou deep intelligence ocr
- Save recent projects path
- Ask if merge boxes if OCR is chosen for Batch translation
- Multiple-box dragging
- Multiple-box alignment
- Add background color, stroke and rotation to global font style
- Add paste image context menu item
- If right2left, append text of the latter one first
- Add mask generator and inpaint plugins support. The first plugin of this kind is Sickzil-Machine
- Add Tesseract’s text lines detection
- Add operations like single-image pretranslation and one-click translation (see the imageviewer’s contextmenu)
- Replace scene text detection with OCR for Batch Translation (or one-click translation)
- Add minimum font size setting
- Add automatic textarea resize option
- Use checkbox instead of button to view translated pictures
- Search and Replace supports replacing source text
- Other minor improvements
- New OCR plugin: easyOCR, tencent
- OCR languages list only shows languages OCR engines support
- Save the geometry to target geometry before expanding
- Fix the localization of TabPane
- Add machine translation for screen reader
- Remove some third-party libraries
- Update OpenCV to 4.5.0
- Support getting text area confidence in offline mode and show progress
- New OCR plugin: WinRT OCR. It uses the built-in OCR API of Windows Runtime on Windows 10. It requires installation of relevant languages on the system
- Some machine translation plugins of BasicCAT are packaged
- UI improvements
- Generate mask and text-removed images according to order and show progress
- Inpainting is now asynchronous
- Validate geometry before generating text mask
- Fix a resized mask not matching with the resized image problem
- Add text area action menu to imagescrollpane
- Maskeditor can now regenerate mask of selected area and scale images
- Text Remover can remove text in the selected area
- The default text mask color is reset to red
- Add text-removed image editor
- Save text mask in png format, making cover text mode supports areas which have a black ground
- New action: Duplicate text area
- Vertical text engine for CJK (Chinese, Japanese and Korean)
- AutoCorrect (useful for solving the full-width punctuations input problem on macOS)
- Toolbar has a new tool: Font
- New project and import pictures actions now share the same saved path.
- Better color picking
- New plugin: OCR plugin. Code is open source: github. PaddleOCR is supported.
- New ocr engine: ABBYY (use ABBYY FineReader’s command line interface, windows only)
- New tool: Screen Reader. It can be used as a screenshot tool and screen captures can get OCRed immediately. Captures can also be appended to ImageTrans’s projects.
- A Toolbar is added to provide more operations like text box splitting, merging and selecting.
- MouseCursors are set correctly for different scenarios.
New option to convert vertically aligned Japanese text to horizontally aligned for better OCR results.
- Add Batch Translation
- Add text areas detection based on scene text detection
- Save paths for choosing files and folders
- Add external inpainter (experimental)
- Add localization support. ImageTrans can now be displayed in two languages: English and Chinese.
- More settings
- RichText support
- XLIFF import and export
- WebP-Converter. Convert JPG/BMP/PNG to WebP and vice versa.
- WebImageHelper. Download all the images from websites or download the image under the mouse cursor based on screen positions. This can be used to translate images in web. Images can be replaced with their target version.
© 2023 BasicCAT ― Powered by Jekyll and Textlog theme