opencv处理图像笔记

末蓝、 2023-05-30 08:47 27阅读 0赞

示例参考[\[OpenCV-Python\] OpenCV 核心操作 部分 III][OpenCV-Python_ OpenCV _ _ III]  
**1 roi**  
提取感兴趣区域，\`\`

import cv2
    import numpy as np
    from matplotlib import pyplot as plt
    # cv2读入的是BGR格式
    img = cv2.imread('test.jpg')
    # 注意H*W，第一个参数为高度,y方向的阈值，第二个参数为宽度，x方向的阈值，比如我这里取眼睛
    r_eye = img[283:292,166:221]
    # matplotlib显示的是RGB，故需要做一下转换
    # 还可以通过cv2.cvtColor(r_eye,cv2.COLOR_BGR2RGB),进行格式转换
    # [2,1,0]更改通道顺序
    plt.imshow(r_eye[:,:,[2,1,0]])
    plt.show()

![2][]  
**2 图像算术运算**  
**2.1 图像混合**

# 图片融合，图片大小必须一致
    ax = fig.add_subplot(245)
    ax.imshow(cv2.cvtColor(bg,cv2.COLOR_BGR2RGB))
    # img.shape[0]是高度，img.shape[1]是宽度
    # cv2.resize中的顺序却是宽度、高度
    bg = cv2.resize(bg,(img.shape[1],img.shape[0]))
    ax = fig.add_subplot(246)
    ax.imshow(cv2.cvtColor(bg,cv2.COLOR_BGR2RGB))
    ax = fig.add_subplot(247)
    img_mer = cv2.addWeighted(img,0.7,bg,0.3,0)
    ax.imshow(cv2.cvtColor(img_mer,cv2.COLOR_BGR2RGB))

![1][]  
**2.2 按位运算**  
[\[Python图像处理\] 十四.基于OpenCV和像素处理的图像灰度化处理][Python_ _._OpenCV]，因为RGB并不能反映物体的形态特征，只是从光学原理上颜色的调配。对文字识别的影响并不大。灰度化后，图片就可以降维了。  
[机器视觉的图像二值化详细分析][Link 1]，最简单的图片分割方法就是二值化，二值图像每个像素只有两种取值：要么纯黑，要么纯白

# 将图片灰度化
    bg_gray = cv2.cvtColor(bg_src,cv2.COLOR_BGR2GRAY)
     # 图片二值化
    ret, mask = cv2.threshold(bg_gray, 10, 255, cv2.THRESH_BINARY)

![1][1 1]  
调整阈值由10调整100后，  
![1][1 2]  
阈值的选取有两种办法，手工调整太费劲，于是就有一些算法产生了。  
一、全局阈值  
指的是对整个图像中的每一个像素都选用相同的阈值。我们可以在Photoshop的图像-调整-阈值里体验,如果采用大津算法,即`ret9, mask9 = cv2.threshold(bg_gray9, 0, 255, cv2.THRESH_OTSU)`,这个时候thresh设置成什么都不影响结果,`ret9`这里返回最佳的阈值

二、局部阈值  
[python-opencv函数总结之（一）threshold、adaptiveThreshold、Otsu 二值化][python-opencv_threshold_adaptiveThreshold_Otsu]，从下图可以看到adaptiveThreshold对于密集型的目标较为合适，他纹理都可以画出来。而主体轮廓反而不需要那么清晰

ax = fig.add_subplot(3,5,11)
    img9 = cv2.imread('9.jpg')
    bg_gray9 = cv2.cvtColor(img9,cv2.COLOR_BGR2GRAY)
    # blockSize必须为奇数，不能为偶数
    # C常量，对画面也有影响
    mask9 = cv2.adaptiveThreshold(bg_gray9,255,cv2.ADAPTIVE_THRESH_MEAN_C,cv2.THRESH_BINARY,21,2)
    ax.imshow(cv2.cvtColor(mask9,cv2.COLOR_BGR2RGB))
    #
    ax = fig.add_subplot(3,5,12)
    ret9, mask9 = cv2.threshold(bg_gray9, 0, 255, cv2.THRESH_OTSU)
    ax.imshow(cv2.cvtColor(mask9,cv2.COLOR_BGR2RGB))
    #
    ax = fig.add_subplot(3,5,14)
    mask9 = cv2.adaptiveThreshold(bg_gray9,255,cv2.ADAPTIVE_THRESH_MEAN_C,cv2.THRESH_BINARY,21,0)
    ax.imshow(cv2.cvtColor(mask9,cv2.COLOR_BGR2RGB))

![2][2 1]  
roi替换,[OpenCV之bitwise\_and、bitwise\_not等图像基本运算及掩膜][OpenCV_bitwise_and_bitwise_not]

bg_src = cv2.imread('test3.jpg')
    ax = fig.add_subplot(3,7,8)
    img = cv2.resize(img,(bg_src.shape[1],img.shape[0]))
    rows,cols,channels = bg_src.shape
    roi = img[0:rows,0:cols]
    bg_gray = cv2.cvtColor(bg_src,cv2.COLOR_BGR2GRAY)
    ax.imshow(cv2.cvtColor(bg_gray,cv2.COLOR_BGR2RGB))
    #
    ax = fig.add_subplot(3,7,9)
    ret, mask = cv2.threshold(bg_gray, 0, 255, cv2.THRESH_OTSU)
    ax.imshow(cv2.cvtColor(mask,cv2.COLOR_BGR2RGB))
    # 
    mask_inv = cv2.bitwise_not(mask)
    img1_bg = cv2.bitwise_and(roi,roi,mask=mask_inv)
    img2_bg = cv2.bitwise_and(bg_src,bg_src,mask=mask)
    dst = cv2.add(img1_bg,img2_bg)
    img[0:rows,0:cols]=dst
    # mask_inv取反后，非mask区域
    ax = fig.add_subplot(3,7,10)
    ax.imshow(cv2.cvtColor(mask_inv,cv2.COLOR_BGR2RGB))
    # 
    ax = fig.add_subplot(3,7,11)
    ax.imshow(cv2.cvtColor(img1_bg,cv2.COLOR_BGR2RGB))
    # 
    ax = fig.add_subplot(3,7,12)
    ax.imshow(cv2.cvtColor(img2_bg,cv2.COLOR_BGR2RGB))
    # 
    ax = fig.add_subplot(3,7,13)
    ax.imshow(cv2.cvtColor(dst,cv2.COLOR_BGR2RGB))
    ax = fig.add_subplot(3,7,14)
    ax.imshow(cv2.cvtColor(img,cv2.COLOR_BGR2RGB))

![3][]  
**3 几何变换**  
**3.1 旋转**  
[opencv中imread第二个参数的含义][opencv_imread]  
[Opencv-Python学习笔记五——图像翻转，平移，仿射及透视 warpAffine][Opencv-Python_ warpAffine]

import cv2
    from matplotlib import pyplot as plt
    
    fig = plt.figure(figsize=(15,15),dpi=60)
    # 读取灰度图
    img = cv2.imread('test3.jpg',0)
    ax = fig.add_subplot(3,4,1)
    ax.imshow(cv2.cvtColor(img,cv2.COLOR_BGR2RGB))
    # 缩放
    h,w = img.shape[:2]
    img2 = cv2.resize(img,(2*w,2*h),interpolation=cv2.INTER_CUBIC)
    ax = fig.add_subplot(3,4,2)
    ax.imshow(cv2.cvtColor(img2,cv2.COLOR_BGR2RGB))
    ax = fig.add_subplot(3,4,3)
    rows,cols = img.shape
    #这里的第一个参数为旋转中心，第二个为旋转角度，第三个为旋转后的缩放因子
    # 可以通过设置旋转中心，缩放因子，以及窗口大小来防止旋转后超出边界的问题
    # M为变换矩阵，在中心点旋转45度，并缩放0.9倍
    M = cv2.getRotationMatrix2D((cols/2,rows/2),45,0.9)
    # 几何变换，同时更改输出图片大小尺寸
    dst = cv2.warpAffine(img,M,(cols,rows*2))
    ax.imshow(cv2.cvtColor(dst,cv2.COLOR_BGR2RGB))
    #
    plt.show()

![4][]  
**3.2 仿射变换**  
图像的旋转加上拉升就是图像仿射变换  
[opencv学习(三十五)之仿射变换warpAffine][opencv_warpAffine]  
[opencv python 图像缩放/图像平移/图像旋转/仿射变换/透视变换][opencv python]  
**3.3 透视变换**  
[透视变换之变换为鸟瞰图][Link 2]  
[透视投影矩阵推导][Link 3]，

img = cv2.imread('test3.jpg')
    ax = fig.add_subplot(3,3,4)
    rows,cols,ch = img.shape
    # 指定一个2*3矩阵，
    pts1 = np.float32([[50,50],[200,50],[50,200]])
    pts2 = np.float32([[10,100],[200,50],[100,250]])
    M = cv2.getAffineTransform(pts1,pts2)
    dst = cv2.warpAffine(img,M,(cols,rows))
    ax.imshow(cv2.cvtColor(dst,cv2.COLOR_BGR2RGB))
    # 透视变换
    ax = fig.add_subplot(3,3,5)
    pts1 = np.float32([[56,65],[368,52],[28,387],[389,390]])
    pts2 = np.float32([[0,0],[300,0],[0,300],[300,300]])
    M = cv2.getPerspectiveTransform(pts1,pts2)
    dst = cv2.warpPerspective(img,M,(cols,rows))
    ax.imshow(cv2.cvtColor(dst,cv2.COLOR_BGR2RGB))
    ax = fig.add_subplot(3,3,6)
    ax.imshow(cv2.cvtColor(img,cv2.COLOR_BGR2RGB))

![5][]  
**4 图片修复**

[OpenCV-Python_ OpenCV _ _ III]: https://www.cnblogs.com/Undo-self-blog/p/8424220.html
[2]: https://img-blog.csdnimg.cn/20191108162313432.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dhcnJhaA==,size_16,color_FFFFFF,t_70
[1]: https://img-blog.csdnimg.cn/2019110816112747.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dhcnJhaA==,size_16,color_FFFFFF,t_70
[Python_ _._OpenCV]: https://blog.csdn.net/Eastmount/article/details/88785768
[Link 1]: http://m.elecfans.com/article/957431.html
[1 1]: https://img-blog.csdnimg.cn/20191111101351843.png
[1 2]: https://img-blog.csdnimg.cn/20191111104655868.png
[python-opencv_threshold_adaptiveThreshold_Otsu]: https://blog.csdn.net/sinat_21258931/article/details/61418681
[2 1]: https://img-blog.csdnimg.cn/20191111114112760.png
[OpenCV_bitwise_and_bitwise_not]: https://blog.csdn.net/u011028345/article/details/77278467
[3]: https://img-blog.csdnimg.cn/20191111115358892.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dhcnJhaA==,size_16,color_FFFFFF,t_70
[opencv_imread]: https://blog.csdn.net/z914022466/article/details/52709981
[Opencv-Python_ warpAffine]: https://www.jianshu.com/p/ef67cacf442c
[4]: https://img-blog.csdnimg.cn/20191111142228217.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dhcnJhaA==,size_16,color_FFFFFF,t_70
[opencv_warpAffine]: https://blog.csdn.net/keith_bb/article/details/56331356
[opencv python]: https://segmentfault.com/a/1190000015645951
[Link 2]: https://www.jianshu.com/p/b49f9dbb26ea
[Link 3]: https://www.cnblogs.com/AirGuanZ/p/6365702.html
[5]: https://img-blog.csdnimg.cn/20191111145908341.png