【持续更新】Python和C++下OpenCV使用小记

OpenCV
—— Open Source Computer Vision

Python下使用示例：

图片篇

0. 导入OpenCV

   import cv2
   

1. 读取图片与写图片

   pic = cv2.imread(pic_path)#bgr格式，shape=(height, width, channel)
   cv2.imwrite(pic_path, pic)
   

   注：cv2.imread返回一个[height, width, channel]的numpy.ndarray对象，height表示图片高度，width表示图片宽度，channel表示图片的通道。

2. 调整尺寸

   pic2 = cv2.resize(pic, (256, 256), interpolation=cv2.INTER_CUBIC)
   

3. 图片分离成多通道，多通道合成图片

   #通道分离
   img_bgra = cv2.imread("sample.png",-1)
   b,g,r,a = cv2.split(img_bgra)
   

   #通道合并
   import numpy as np
   b_channel = np.ones((256, 256), dtype=np.uint8) * 0
   g_channel = b_channel
   r_channel = b_channel
   image_bgr = cv2.merge((b_channel, g_channel, r_channel))
   

4. 将三通道图转为单通道图（灰度图）

   im_gray = cv2.cvtColor(image,cv2.COLOR_RGB2GRAY)
   

5. 在图片上画圆/点

   center = (int(x), int(y))
   radius = 1
   color = [0, 255, 255]
   cv2.circle(img_bgr, center, radius, color, -1)
   

6. 图片仿射

   cv2.warpAffine(src, affine_mat, size, dst, cv2.WARP_INVERSE_MAP, borderMode=cv2.BORDER_TRANSPARENT)
   

7. 图片边界填充

   output_bgr = cv2.copyMakeBorder(input_bgr, 1, 1, 1, 1, cv2.BORDER_ISOLATED | cv2.BORDER_CONSTANT, value=0)#用0填充上下左右边界1像素边界
   

8. 返回最小的外接矩形

   l, t, w, h = cv2.boundingRect(mask)#l为x起点，w为矩形宽(横向)；t为y起点，h为矩形高（纵向）
   

9. 在图片上画矩形

   cv2.rectangle(img_bgr, (int(l), int(t)),(int(l+w), int(t+h)), (0, 255, 0), 1)
   

10. 腐蚀/膨胀

    erode_mask = cv2.erode(mask, cv2.getStructuringElement( cv2.MORPH_ELLIPSE, (3, 3)), iterations=1)
    dilate_mask = cv2.dilate(mask, cv2.getStructuringElement( cv2.MORPH_ELLIPSE, (3, 3)), iterations=1)
    

11. 高斯模糊

    blur_mask = cv2.GaussianBlur(mask, (3, 3), 0)
    

12. alpha融合
    将两张图片src1和src2进行alpha融合（加权叠加）。

    dst=α⋅src1+β⋅src2+γ
    

    src1 = cv2.imread(cv2.samples.findFile('LinuxLogo.jpg'))
    src2 = cv2.imread(cv2.samples.findFile('WindowsLogo.jpg'))
    alpha = 0.5#所有像素值均乘以0.5
    beta = (1.0 - alpha)
    gamma = 0.0
    dst = cv2.addWeighted(src1, alpha, src2, beta, gamma)
    

    详细使用：Adding (blending) two images using OpenCV

    进阶：cv2.multiply( )，根据图片自带alpha蒙版融合

    src1 = cv2.imread('src1.png',-1)#读取bgra图片，alpha通道为自带蒙版（非上面固定的alpha = 0.5）
    src2 = cv2.imread('src2.jpg')#读取bgr图片
    
    #通道分离
    b,g,r,a = cv2.split(src1)
    
    #得到src1前景部分
    src1_foreground = cv2.merge((b,g,r))
    
    #得到src1的alpha蒙版
    src1_alpha = cv2.merge((a,a,a))
    
    #因为下面要进行乘法运算故将数据类型设为float，防止溢出
    src1_foreground = src1_foreground.astype(float)
    src2 = src2.astype(float)
    
    #将alpha的值归一化在0-1之间，作为加权系数
    src1_alpha = src1_alpha.astype(float)/255
    
    #加权，每个像素的加权系数即为alpha蒙版对应位置像素的值
    src1_foreground = cv2.multiply(src1_alpha,src1_foreground)
    src2 = cv2.multiply(1-src1_alpha,src2)
    outImage = src1_foreground + src2
    

    参考：LearnOpencv进阶系列一之Alpha通道混合

13. 图片拼接（需图片尺寸相同）

im = cv2.imread(pic_path)
im_hconcat = cv2.hconcat([im, im]) # 水平拼接，个数可以多个
im_vconcat = cv2.vconcat([im, im, im]) # 垂直拼接

视频篇

OpenCV集成了较完善的视频处理接口，不过由于OpenCV重点在于图像处理，因此可能不提供视频处理中某些常用的功能，例如视频生成的同时合并音频。如需这些视频处理的高级功能，建议使用FFmpeg。

1. 读取视频文件

   def cv2_read_video(video_file_path):
       """支持读取视频和GIF"""
       start = time.time()
       cap = cv2.VideoCapture(file_path)
       frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))#帧宽，横向
       frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))#帧高，纵向
       fps = cap.get(cv2.CAP_PROP_FPS)
       res = []
       while True:
          ret, frame = cap.read()
          if ret is False:
              break
          res.append(frame)
       total_frames = len(res)
       cap.release()
       end = time.time()
       print('the video has total frames: ',total_frames)
       print('opencv read video use time: '+str(end-start))
       return res, frame_width, frame_height, fps
   

   Windows平台下还支持使用GPU解码读取视频：

   import cv2
   import time
   
   def load_material(file_path):
       """
       Load a full list of frames in memory.
       """
       cap = cv2.VideoCapture(file_path)
       frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
       frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
       fps = cap.get(cv2.CAP_PROP_FPS)
       res = []
       while True:
           ret, frame = cap.read()
           if ret is False:
               break
           res.append(frame)
       material = res
       total_frames = len(res)
       cap.release()
       return material
   
   def load_material_cuda(file_path):
       """
       Load a full list of frames in memory using cuda.
       """
       cap = cv2.cudacodec.createVideoReader(file_path)  # only support on Windows
       res = []
       while True:
           ret, frame = cap.nextFrame()
           if ret is False:
               break
           res.append(frame.download())
       material = res
       total_frames = len(res)
       cap.release()
       return material
   
   def main(file_path):
       t1 = time.time()
       material = load_material(file_path)
       t2 = time.time()
       print('opencv use cpu load material use: ', t2-t1)
       material2 = load_material_cuda(file_path)
       t3 = time.time()
       print('opencv use gpu load material use: ', t3-t2)
   
   if __name__ == '__main__':
       file_path = 'test.mp4'
       main(file_path)
       
   

2. 写视频文件

   def cv2_output_no_audio_video(output_imgs, output_file_path, fps, frame_size):
       """注：输出视频无音频"""
       size = frame_size
       start = time.time()
       fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
       vout = cv2.VideoWriter()
       vout.open(output_file_path, fourcc, fps, size, True)
       for img in output_imgs:
           vout.write(img)
       vout.release()
       end = time.time()
       print('cv2_output_no_audio_video use time: '+str(end-start))
   

C++下使用示例：

图片篇

1. 获取Mat的最大最小值

   double minVal, maxVal;
   cv::minMaxIdx(compare_binaryMaskFloat, &minVal, &maxVal); 
   

2. 抽取多个通道（rgb）图片中的某通道

   int coi = 0;//1, 2
   cv::extractChannel(src, dst, coi)	
   

参考资料：

[1] OpenCV-Python Tutorials
[2] LearnOpencv进阶系列一之Alpha通道混合
[3] OpenCV concat
[4] github - OpenCV-Python-Tutorial
[5] OpenCV-Python中文教程 · 看云
[6] FFmpeg常用命令小结
[7] OpenCV: Video I/O with OpenCV Overview
[8] opencv 颜色空间转换函数cvtColor()运用
[9] void cv::cuda::cvtColor
[10] void cv::cvtColor
[11] OpenCV: cv::VideoCapture Class Reference
[12] Learn OpenCV ( C++ / Python )
[13] void cv::extractChannel