我正在尝试为用 C/C++ 编写的应用程序创建一个 python 包装器，它广泛使用 OpenCV C API。我想为此使用ctypes，因为我已经在以前的程序中成功使用过它。但是当我尝试将 Python 中的 IplImage 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：
// 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：
// 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：
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cv2.cv 模块中的 python 绑定替换最后一行image_show(pointer(cv_img))时：
cv.ShowImage("image", cv_img)
cv.WaitKey(20)
然后我得到正确的输出。
所以问题在于image_show(pointer(cv_img))失败并出现TypeError: 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：
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：
# 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)]
视频循环然后更改为
# 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，但随后它为OpenCV Error: Bad flag (parameter or structure field) (Unrecognized or unsupported array type) in unknown function哭泣。所以创建的 ctypes 结构是无效的。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，我什至会接受不使用 ctypes 的其他解决方案。谢谢。
注意：我在过去 2 天里试图找到这个问题的答案，但没有成功。OpenCV 1.0 有一些解决方案，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。:(