关于opencv:使用ctypes将python IplImage对象作为简单结构传递给共享C库

Pass python IplImage object as simple struct to shared C library using ctypes

我正在尝试为用 C/C 编写的应用程序创建一个 python package器,它广泛使用了 OpenCV C API。我想为此使用 ctypes,因为我在以前的程序中成功使用了它。但是当我尝试将 IplImage 从 Python 作为参数传递给 c 库中的函数时遇到了问题。

我创建了一个示例测试库来演示该问题。以下是我想使用的库中的函数:

1
2
3
4
5
6
7
8
9
10
11
12
// ImageDll.h

#include"opencv2/opencv.hpp"

extern"C"  //Tells the compile to use C-linkage for the next scope.
{
    // Returns image loaded from location
    __declspec(dllexport) IplImage* Load(char* dir);

    // Show image
    __declspec(dllexport) void Show(IplImage* img);
}

还有一个cpp文件:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
// ImageDll.cpp
// compile with: /EHsc /LD

#include"ImageDll.h"

using namespace std;

extern"C"  //Tells the compile to use C-linkage for the next scope.
{
    IplImage* Load(char* dir)
    {
        return cvLoadImage(dir, CV_LOAD_IMAGE_COLOR);
    }

    void Show(IplImage* img)
    {
        cvShowImage("image", img);
        cvWaitKey(0);
    }
}

这是python的初步尝试:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
from time import sleep
from ctypes import *
from modules.acquisition import InitCamera, GetImage
from modules.utils import struct
import cv2.cv as cv

# load DLL containing image functions
print"Loading shared library with genetic algorithm...",
image_lib = cdll.LoadLibrary("OpenCV_test_DLL.dll")
print"Done."
# get function handles
print"Loading functions of library...",
image_load = image_lib.Load
image_show = image_lib.Show
# set return type for functions (because ctypes default is int)
image_load.restype = c_void_p
image_show.restype = None
print"Done."

# initialize source
print"Initializing camera",
source = struct()
InitCamera(source)
print"Done."

# show video
while (1):
    # get image as PIL image
    img = GetImage(source)
    # transform image to OpenCV IplImage
    cv_img = cv.CreateImageHeader(img.size, cv.IPL_DEPTH_8U, 3)
    cv.SetData(cv_img, img.tostring())
    # show image using OpenCV highgui lib
    image_show(pointer(cv_img))

如您所见,我从相机获取图像作为 PIL 图像,然后将其转换为 python IplImage。这适用于 100%,因为当我用来自 cv2.cv 模块的 python 绑定替换最后一行 image_show(pointer(cv_img)) 时:

1
2
    cv.ShowImage("image", cv_img)
    cv.WaitKey(20)

然后我得到正确的输出。

所以问题出在 image_show(pointer(cv_img)) 上,它因 TypeError 失败:type must have storage info。这是因为 cv_img 需要是有效的 ctypes IplImage 结构。我试图用 ctypes 模仿它,但收效甚微:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
from ctypes import *
from cv2 import cv

# ctypes IplImage
class cIplImage(Structure):
    _fields_ = [("nSize", c_int),
                ("ID", c_int),
                ("nChannels", c_int),
                ("alphaChannel", c_int),
                ("depth", c_int),
                ("colorModel", c_char * 4),
                ("channelSeq", c_char * 4),
                ("dataOrder", c_int),
                ("origin", c_int),
                ("align", c_int),
                ("width", c_int),
                ("height", c_int),
                ("roi", c_void_p),
                ("maskROI", c_void_p),
                ("imageID", c_void_p),
                ("tileInfo", c_void_p),
                ("imageSize", c_int),
                ("imageData", c_char_p),
                ("widthStep", c_int),
                ("BorderMode", c_int * 4),
                ("BorderConst", c_int * 4),
                ("imageDataOrigin", c_char_p)]

这是进行转换的函数:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
# convert Python PIL to ctypes Ipl
def PIL2Ipl(input_img):

    # mode dictionary:
    # (pil_mode : (ipl_depth, ipl_channels)
    mode_list = {
       "RGB" : (cv.IPL_DEPTH_8U, 3),
       "L"   : (cv.IPL_DEPTH_8U, 1),
       "F"   : (cv.IPL_DEPTH_32F, 1)
        }

    if not mode_list.has_key(input_img.mode):
        raise ValueError, 'unknown or unsupported input mode'

    result = cIplImage()
    result.imageData = c_char_p(input_img.tostring())
    result.depth = c_int(mode_list[input_img.mode][0])
    result.channels = c_int(mode_list[input_img.mode][1])
    result.height = c_int(input_img.size[0])
    result.width = c_int(input_img.size[1])

    return result
                    ("imageData", c_char_p),
                    ("widthStep", c_int),
                    ("BorderMode", c_int * 4),
                    ("BorderConst", c_int * 4),
                    ("imageDataOrigin", c_char_p)]

视频循环然后变为

1
2
3
4
5
6
7
8
9
# show video
while (1):
    # get image as PIL image
    img = GetImage(source)
    # transform image to OpenCV IplImage
    cv_img = cIplImage()
    cv_img = PIL2Ipl(img)
    # show image using OpenCV highgui lib
    image_show(pointer(cv_img))

通过这种方式,数据被传递到库,但随后它在未知函数中为 OpenCV 错误而哭泣:错误标志(参数或结构字段)(无法识别或不支持的数组类型)。因此创建的 ctypes 结构无效。有谁知道如何正确实施它?当它使我能够将 python IplImage 传递给 c 库时,我什至会接受其他不使用 ctypes 的解决方案。谢谢。

注意:过去 2 天我一直试图找到这个问题的答案,但没有任何成功。有针对 OpenCV 1.0 的解决方案,但最近使用 numpy 数组的 Python 的 OpenCV 绑定几乎不可能使 Python 和 C 应用程序之间的接口正常工作。 :(


我终于找到了解决问题的方法。而不是使用默认的python函数

1
2
cv.CreateImageHeader()
cv.SetData()

我使用了从 OpenCV C 库导出的 C 函数。 :) 我什至设法将颜色从 PILs RGB 转换为 IplImage BGR 格式。这是完整的来源:

ImageDll.h

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
#include"opencv2/opencv.hpp"

extern"C"  //Tells the compile to use C-linkage for the next scope.
{
    // Returns image loaded from location
    __declspec(dllexport) IplImage* Load(char* dir);

    // Show image
    __declspec(dllexport) void Show(IplImage* img);

    // Auxiliary functions
    __declspec(dllexport) void aux_cvSetData(CvArr* arr, void* data, int step);
    __declspec(dllexport) IplImage* aux_cvCreateImageHeader(int width, int height, int depth, int channels);
    __declspec(dllexport) IplImage* aux_cvCvtColor(const IplImage* src, int code);
    __declspec(dllexport) void aux_cvCopy(const CvArr* src, CvArr* dst);
    __declspec(dllexport) void aux_cvReleaseImage(IplImage** image);
    __declspec(dllexport) void aux_cvReleaseImageHeader(IplImage** image);
}

ImageDll.cpp

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
#include"ImageDll.h"

using namespace std;

extern"C"  //Tells the compile to use C-linkage for the next scope.
{
    IplImage* Load(char* dir)
    {
        return cvLoadImage(dir, CV_LOAD_IMAGE_COLOR);
    }

    void Show(IplImage* img)
    {
        cvShowImage("image", img);
        cvWaitKey(5);
    }

    void aux_cvSetData(CvArr* arr, void* data, int step)
    {
        cvSetData(arr,data,step);
    }

    IplImage* aux_cvCreateImageHeader(int width, int height, int depth, int channels)
    {
        return cvCreateImageHeader(cvSize(width,height), depth, channels);
    }

    IplImage* aux_cvCvtColor(const IplImage* src, int code)
    {
        IplImage* dst = cvCreateImage(cvSize(src->width,src->height),src->depth,src->nChannels);
        cvCvtColor(src, dst, code);
        return dst;
    }

    void aux_cvCopy(const CvArr* src, CvArr* dst)
    {
        cvCopy(src, dst, NULL);
    }

    void aux_cvReleaseImage(IplImage** image)
    {
        cvReleaseImage(image);
    }

    void aux_cvReleaseImageHeader(IplImage** image)
    {
        cvReleaseImageHeader(image);
    }
}

run.py

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
# This Python file uses the following encoding: utf-8

from time import sleep
from ctypes import *
from modules.acquisition import InitCamera, GetImage
from modules.utils import struct
import cv2.cv as cv
from modules.ipl import *

# load DLL containing image functions
print"Loading shared library with C functions...",
image_lib = cdll.LoadLibrary("OpenCV_test_DLL.dll")
print"Done."
# get function handles
print"Loading functions of library...",
image_load = image_lib.Load
image_show = image_lib.Show
cvReleaseImage = image_lib.aux_cvReleaseImage
# set return type for functions (because ctypes default is int)
image_load.restype = c_void_p
image_show.restype = None
cvReleaseImage.restype = None
print"Done."

# initialize source
print"Initializing camera",
source = struct()
InitCamera(source)
print"Done."

# show video
while (1):
    # get image as PIL image
    img = GetImage(source)
    # transform image to OpenCV IplImage
    cv_img = PIL2Ipl(img)
    # show image using OpenCV highgui lib
    image_show(cv_img)
    # release memory
    cvReleaseImage(byref(cv_img))

ipl.py

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
from ctypes import *
from cv2 import cv

# ctypes IplImage
class cIplImage(Structure):
    _fields_ = [("nSize", c_int),
                ("ID", c_int),
                ("nChannels", c_int),
                ("alphaChannel", c_int),
                ("depth", c_int),
                ("colorModel", c_char * 4),
                ("channelSeq", c_char * 4),
                ("dataOrder", c_int),
                ("origin", c_int),
                ("align", c_int),
                ("width", c_int),
                ("height", c_int),
                ("roi", c_void_p),
                ("maskROI", c_void_p),
                ("imageID", c_void_p),
                ("tileInfo", c_void_p),
                ("imageSize", c_int),
                ("imageData", POINTER(c_char)),
                ("widthStep", c_int),
                ("BorderMode", c_int * 4),
                ("BorderConst", c_int * 4),
                ("imageDataOrigin", c_char_p)]

# load DLL containing needed OpenCV functions
libr = cdll.LoadLibrary("OpenCV_test_DLL.dll")
cvSetData = libr.aux_cvSetData
cvCreateImageHeader = libr.aux_cvCreateImageHeader
cvCvtColor = libr.aux_cvCvtColor
cvCopy = libr.aux_cvCopy
cvReleaseImage = libr.aux_cvReleaseImage
cvReleaseImageHeader = libr.aux_cvReleaseImageHeader
# set return types for library functions
cvSetData.restype = None
cvCreateImageHeader.restype = POINTER(cIplImage)
cvCvtColor.restype = POINTER(cIplImage)
cvCopy.restype = None
cvReleaseImage.restype = None
cvReleaseImageHeader.restype = None
#print"auxlib loaded"

# convert Python PIL to ctypes Ipl
def PIL2Ipl(pil_img):
   """Converts a PIL image to the OpenCV/IplImage data format.

    Supported input image formats are:
        RGB
        L
        F
   """


    # mode dictionary:
    # (pil_mode : (ipl_depth, ipl_channels)
    mode_list = {
       "RGB" : (cv.IPL_DEPTH_8U, 3),
       "L"   : (cv.IPL_DEPTH_8U, 1),
       "F"   : (cv.IPL_DEPTH_32F, 1)
        }

    if not mode_list.has_key(pil_img.mode):
        raise ValueError, 'unknown or unsupported input mode'

    depth = c_int(mode_list[pil_img.mode][0])
    channels = c_int(mode_list[pil_img.mode][1])
    height = c_int(pil_img.size[1])
    width = c_int(pil_img.size[0])
    data = pil_img.tostring()

    ipl_img = cvCreateImageHeader(width, height, depth, channels);
    cvSetData(ipl_img, create_string_buffer(data,len(data)), c_int(width.value * channels.value))
    brg_img = cvCvtColor(ipl_img,cv.CV_RGB2BGR)
    cvReleaseImageHeader(byref(ipl_img))
    return brg_img

希望对大家有所帮助:)