naptha/tesseract.js-core

tesseract.js-core

Core part of tesseract.js, which compiles original tesseract from C to JavaScript WebAssembly.

Compiling

To build tesseract-core.js by yourself, please install docker and run:

bash build-with-docker.sh

The generated files will be stored in root path. When compiling, errors sometimes occur due to race conditions (some dependencies do not appear to compile properly in parallel). Re-running generally resolves.

Structure

1. Build scripts are in build-scripts folder
2. Javascript/wrapper files are in javascript folder
3. All dependencies (including Tesseract) are in third_party folder
   1. All dependencies are unmodified except for Tesseract, which uses a forked repo
   2. The Tesseract repo has the following changes:
      1. Modified CMakeLists.txt to build with emscripten
      2. Modified ltrresultiterator.h and ltrresultiterator.cpp to add WordChoiceIterator class
      3. Added src/arch_sse folder containing copy of src/arch with various changes for the SIMD builds.
         1. Hard-coded use of the SSE function because Tesseract feature detection does not work in Webassembly and made various performance optimizations (in src/arch_sse/intsimdmatrixsse.cpp).
         2. Modified src/arch_sse/intsimdmatrixavx2.cpp and src/arch_sse/simddetect.cpp for Relaxed SIMD build.
      4. Commented out "Empty page!!" message in src/textord/colfind.cpp to prevent this from printing to console
      5. Added functions for detecting page angle and applying rotation
         1. Modified src/ccmain/thresholder.cpp, src/ccmain/thresholder.h, src/api/baseapi.cpp, and include/tesseract/baseapi.h to add exif and angle arguments for rotating images
         2. Changed FindLines from "protected" to "public" in baseapi.h to expose to Javascript
            1. Allows for lines (and therefore page angle) to be detected without running unnecessary steps afterwards
         3. Added public GetGradient function to baseapi.h and baseapi.cpp for reporting page angle
            1. Also required minor changes to src/ccmain/tesseractclass.h, src/ccmain/pagesegmain.cpp, src/textord/textord.cpp, and src/textord/textord.h
               1. See this commit: https://github.com/Balearica/tesseract/commit/db6951f655263878f6344380a5ddb95e678d7c09
      6. Added WriteImage function to baseapi.h and baseapi.cpp for saving images (original, grey, and binary)
      7. Added SaveParameters and RestoreParameters functions to baseapi.h and baseapi.cpp for saving and restoring parameters
      8. Added calls to EM_ASM_ARGS to src/ccmain/control.cpp for progress logging (and added <emscripten.h> header)
      9. Rewrote tprintf function in src/ccutil/tprintf.cpp to force flushing
      10. Added new version of SetImage to src/api/baseapi.cpp and include/tesseract/baseapi.h that reads image from filesystem
          1. This was done to resolve memory leak--see this issue
      11. Edited ParamUtils::PrintParams in src/ccutil/params.cpp to remove description text (resolves bug)
          1. The bug was reported in this Git Issue, so we can cut this point if resolved in a future version of Tesseract
      12. Edited src/ccmain/tessedit.cpp to save error log to separate file (/debugDev.txt)
      13. Added JSON as an ouput format
          1. Added src/api/jsonrenderer.cpp, modified CMakeLists.txt, include/tesseract/baseapi.h, and include/tesseract/renderer.h

Running Minimal Examples

To run the browser examples, launch a web server in the root of the repo (i.e. run http-server). Then navigate to the pages in examples/web/minimal/ in your browser.

To run the node examples, navigate to examples/node/minimal/ and then run e.g. node index.wasm.js [input_file].

The "benchmark" examples behave similarly, except that they take longer to run and report runtime instead of recognition text. All other examples are experimental and should not be expected to run.

Contribution

As we leverage git-submodule to manage dependencies, remember to add recursive when cloning the repository:

git clone --recursive https://github.com/naptha/tesseract.js-core