Using C API of tesseract 3.02 with ctypes and cv2 in python

ctypes, opencv, python, tesseract

Solution

The default `restype` is `c_int`, and the default argument conversion from an integer is also `c_int`. You'll find examples on the web that assume a 32-bit platform that has `sizeof(int) == sizeof(void *)`. This was never a good assumption to make. To protect a 64-bit pointer from truncation when converted to and from a Python integer, set the function pointer's `argtypes` and `restype`. It's a good idea to do this anyway, since it allows ctypes to raise an `ArgumentError` when the the wrong types or number of arguments are used.

If you'd rather not define the prototypes for every function, then at least set `TessBaseAPICreate.restype` to an opaque pointer type.

The following ctypes definitions are based on the header api/capi.h. For convenience I've packaged the API into a `Tesseract` class.

import sys
import cv2
import ctypes
import ctypes.util

if sys.platform == 'win32':
    LIBNAME = 'libtesseract302'
else:
    LIBNAME = 'tesseract'

class TesseractError(Exception):
    pass

class Tesseract(object):
    _lib = None
    _api = None

    class TessBaseAPI(ctypes._Pointer):
        _type_ = type('_TessBaseAPI', (ctypes.Structure,), {})

    @classmethod
    def setup_lib(cls, lib_path=None):
        if cls._lib is not None:
            return
        if lib_path is None:
            lib_path = ctypes.util.find_library(LIBNAME)
            if lib_path is None:
                 raise TesseractError('tesseract library not found')
        cls._lib = lib = ctypes.CDLL(lib_path)

        # source:
        # https://github.com/tesseract-ocr/tesseract/
        #         blob/3.02.02/api/capi.h

        lib.TessBaseAPICreate.restype = cls.TessBaseAPI

        lib.TessBaseAPIDelete.restype = None # void
        lib.TessBaseAPIDelete.argtypes = (
            cls.TessBaseAPI,) # handle

        lib.TessBaseAPIInit3.argtypes = (
            cls.TessBaseAPI, # handle
            ctypes.c_char_p, # datapath
            ctypes.c_char_p) # language

        lib.TessBaseAPISetImage.restype = None
        lib.TessBaseAPISetImage.argtypes = (
            cls.TessBaseAPI, # handle
            ctypes.c_void_p, # imagedata
            ctypes.c_int,    # width
            ctypes.c_int,    # height
            ctypes.c_int,    # bytes_per_pixel
            ctypes.c_int)    # bytes_per_line

        lib.TessBaseAPIGetUTF8Text.restype = ctypes.c_char_p
        lib.TessBaseAPIGetUTF8Text.argtypes = (
            cls.TessBaseAPI,) # handle

    def __init__(self, language='eng', datapath=None, lib_path=None):
        if self._lib is None:
            self.setup_lib(lib_path)
        self._api = self._lib.TessBaseAPICreate()
        if self._lib.TessBaseAPIInit3(self._api, datapath, language):
            raise TesseractError('initialization failed')

    def __del__(self):
        if not self._lib or not self._api:
            return
        if not getattr(self, 'closed', False):
            self._lib.TessBaseAPIDelete(self._api)
            self.closed = True

    def _check_setup(self):
        if not self._lib:
            raise TesseractError('lib not configured')
        if not self._api:
            raise TesseractError('api not created')

    def set_image(self, imagedata, width, height,
                  bytes_per_pixel, bytes_per_line=None):
        self._check_setup()
        if bytes_per_line is None:
            bytes_per_line = width * bytes_per_pixel
        self._lib.TessBaseAPISetImage(self._api,
                                      imagedata, width, height,
                                      bytes_per_pixel, bytes_per_line)

    def get_utf8_text(self):
        self._check_setup()
        return self._lib.TessBaseAPIGetUTF8Text(self._api)

    def get_text(self):
        self._check_setup()
        result = self._lib.TessBaseAPIGetUTF8Text(self._api)
        if result:
            return result.decode('utf-8')

Example usage:

if __name__ == '__main__':
    imcv = cv2.imread('ocrtest.png')
    height, width, depth = imcv.shape

    tess = Tesseract()
    tess.set_image(imcv.ctypes, width, height, depth)
    text = tess.get_text()

    print text.strip()

I tested this on Linux with libtesseract.so.3. Note that `cv2.imread` returns a NumPy array. This has a `ctypes` attribute that includes the `_as_parameter_` hook, set as a `c_void_p` pointer to the the array. Note also that the code shown in the question has the width and height transposed. It should have been `h, w, d = imcv.shape`.

ocrtest.png:

Output:

I am trying to use Tesseract 3.02 with ctypes and cv2 in python. Tesseract
provides a DLL exposed set of C style APIs, one of them is as following:

Problem

Original source