一. 背景

1. 用pyhton可直接调用C++，减少重写的工作量；
2. 部分逻辑运算，C++的执行效率高，可进行加速。
   下面就一个简单的C++滤镜（彩色图转灰度图）为例，展示python调用C++

二. 代码实现

代码结构如下：

.
├── build
├── CMakeLists.txt
├── image_processing.cpp    # C++头文件
├── image_processing.h      # C++源文件
└── image_process.py        # python调用C++

各个文件的内容如下
image_processing.h

#ifndef IMAGE_PROCESSING_H
#define IMAGE_PROCESSING_H

#include <opencv2/opencv.hpp>

extern "C" {
    void process_image(const unsigned char* input, unsigned char* output, int width, int height, int channels);
}

#endif // IMAGE_PROCESSING_H

image_processing.cpp

#include "image_processing.h"

void process_image(const unsigned char* input, unsigned char* output, int width, int height, int channels) {
    cv::Mat input_image(height, width, channels == 3 ? CV_8UC3 : CV_8UC1, (void*)input);
    cv::Mat output_image(height, width, CV_8UC1);

    // 转换为灰度图像
    cv::cvtColor(input_image, output_image, cv::COLOR_BGR2GRAY);

    // 将处理后的图像数据复制到输出缓冲区
    std::memcpy(output, output_image.data, width * height * sizeof(unsigned char));
}

image_process.py

import ctypes
import numpy as np
import cv2
import os

# 确定库文件路径
libname = "./build/libimage_processing.so"

# 加载共享库
image_lib = ctypes.CDLL(libname)

# 定义处理函数的原型
image_lib.process_image.argtypes = [
    ctypes.POINTER(ctypes.c_ubyte),  # 输入图像数据
    ctypes.POINTER(ctypes.c_ubyte),  # 输出图像数据
    ctypes.c_int,                    # 宽度
    ctypes.c_int,                    # 高度
    ctypes.c_int                     # 通道数
]

# 读取图像
input_image = cv2.imread('input.jpg')
height, width, channels = input_image.shape

# 创建输出缓冲区
output_image = np.zeros((height, width), dtype=np.uint8)

# 调用 C++ 处理函数
input_ptr = input_image.ctypes.data_as(ctypes.POINTER(ctypes.c_ubyte))
output_ptr = output_image.ctypes.data_as(ctypes.POINTER(ctypes.c_ubyte))
image_lib.process_image(input_ptr, output_ptr, width, height, channels)

cv2.imwrite("output.jpg", output_image)

CMakeLists.txt

cmake_minimum_required(VERSION 3.10)
project(ImageProcessingLibrary)

# 设置 C++ 标准
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED True)

# 查找 OpenCV 库
find_package(OpenCV REQUIRED)

# 包含 OpenCV 头文件
include_directories(${OpenCV_INCLUDE_DIRS})

# 添加库
add_library(image_processing SHARED image_processing.cpp)

# 链接 OpenCV 库
target_link_libraries(image_processing ${OpenCV_LIBS})

三. 编译代码 && 调用动态库

1. 编译代码

# 执行下面命令后，会生成动态库，./build/libimage_processing.so
mkdir build
cd build
cmake ..
make

2. 调用动态库

python image_process.py

效果如下（模拟灰度滤镜）：