devbox
2023面试高手
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

OpenCV 截取指定区域、黑化背景、透视转换_cv2.warpaffine

作者：2023面试高手 | 2024-06-06 00:08:55

赞

踩

cv2.warpaffine

 

一、cv2.getPerspectiveTransform

cv2.getPerspectiveTransform(src, dst) → retval

src：源图像中待测矩形的四点坐标

sdt：目标图像中矩形的四点坐标

 

一、cv2.warpAffine

放射变换函数，可实现旋转，平移，缩放；变换后的平行线依旧平

cv2.warpAffine(src, M, dsize, dst=None, flags=None, borderMode=None, borderValue=None) --> dst

src：输入图像     dst：输出图像

M：2×3的变换矩阵

dsize：变换后输出图像尺寸

flag：插值方法

borderMode：边界像素外扩方式

borderValue：边界像素插值，默认用0填充

变换矩阵M可通过cv2.getAffineTransfrom(points1, points2)函数获得

变换矩阵的获取需要至少三组变换前后对应的点坐标，设取原图上的三个点组成矩阵points1，变换后的三个点组成的矩阵points2

  1. points1 = np.float32([ [30,30], [100,40], [40,100] ])
  2. points2 = np.float32([ [60,60], [40,100], [80,20] ])
  3.  
  4. M  =  cv2.getAffineTransform(points1, points2)
  5.    =  array([[-0.33333333,  0.33333333, 60.   ],
  6.              [ 0.66666667, -0.66666667, 60.   ]])
  7.  
  8. Affine_img = cv2.warpAffine(img, M, (img.shape[1], img.shape[0])) 

 

二、cv2.warpPerspective

透视变换函数，可保持直线不变形，但是平行线可能不再平行

cv2.warpPerspective(src, M, dsize, dst=None, flags=None, borderMode=None, borderValue=None) --> dst

其相关参数和cv2.warpAffine函数的类似，不再做介绍

它的变换矩阵可以通过cv2.getPerspectiveTransform()函数获得，其原理和cv2.getAffineTransfrom()相同，只是投射变换至少需要四组变换前后对应的点坐标，设取原图上的四个点组成矩阵points1，变换后的四个点组成的矩阵points2

如：

  1. points1 = np.float32([ [30,30], [10,40], [40,10], [5,15] ])
  2. points2 = np.float32([ [0,0], [400,0], [0,400], [400,400] ])
  3.  
  4. M = cv2.getPerspectiveTransform(points1, points2)
  5.   = array([[-9.08777969e+00, -4.54388985e+00,  4.08950086e+02],
  6.            [-5.37005164e+00, -1.07401033e+01,  4.83304647e+02],
  7.            [-1.15318417e-02, -1.35972461e-02,  1.00000000e+00]])
  8.  
  9. Perspective_img = cv2.warpPerspective(img, M, (img.shape[1], img.shape[0])) 

 

三、快速应用工具

  1. # -*- coding: utf-8 -*-
  2. """
  3. @author: yuki_ho
  4. """
  5.  
  6. import cv2
  7. import numpy as np
  8.  
  9. # -----------------------鼠标操作相关------------------------------------------
  10. lsPointsChoose = []
  11. tpPointsChoose = []
  12. pointsCount = 0
  13. count = 0
  14. pointsMax = 6
  15.  
  16.  
  17. def on_mouse(event, x, y, flags, param):
  18.     global img, point1, point2, count, pointsMax
  19.     global lsPointsChoose, tpPointsChoose  # 存入选择的点
  20.     global pointsCount  # 对鼠标按下的点计数
  21.     global img2, ROI_bymouse_flag
  22.     img2 = img.copy()  # 此行代码保证每次都重新再原图画  避免画多了
  23.     # -----------------------------------------------------------
  24.     #    count=count+1
  25.     #    print("callback_count",count)
  26.     # --------------------------------------------------------------
  27.  
  28.     if event == cv2.EVENT_LBUTTONDOWN:  # 左键点击
  29.         pointsCount = pointsCount + 1
  30.         # 为了保存绘制的区域，画的点稍晚清零
  31.         # if (pointsCount == pointsMax + 1):
  32.         #     pointsCount = 0
  33.         #     tpPointsChoose = []
  34.         print('pointsCount:', pointsCount)
  35.         point1 = (x, y)
  36.         print(x, y)
  37.         # 画出点击的点
  38.         cv2.circle(img2, point1, 10, (0, 255, 0), 2)
  39.  
  40.         # 将选取的点保存到list列表里
  41.         lsPointsChoose.append([x, y])  # 用于转化为darry 提取多边形ROI
  42.         tpPointsChoose.append((x, y))  # 用于画点
  43.         # ----------------------------------------------------------------------
  44.         # 将鼠标选的点用直线连起来
  45.         print(len(tpPointsChoose))
  46.         for i in range(len(tpPointsChoose) - 1):
  47.             print('i', i)
  48.             cv2.line(img2, tpPointsChoose[i], tpPointsChoose[i + 1], (0, 0, 255), 2)
  49.         # ----------------------------------------------------------------------
  50.         # ----------点击到pointMax时可以提取去绘图----------------
  51.  
  52.         cv2.imshow('src', img2)
  53.  
  54.     # -------------------------右键按下清除轨迹-----------------------------
  55.     if event == cv2.EVENT_RBUTTONDOWN:  # 右键点击
  56.         print("right-mouse")
  57.         pointsCount = 0
  58.         tpPointsChoose = []
  59.         lsPointsChoose = []
  60.         print(len(tpPointsChoose))
  61.         for i in range(len(tpPointsChoose) - 1):
  62.             print('i', i)
  63.             cv2.line(img2, tpPointsChoose[i], tpPointsChoose[i + 1], (0, 0, 255), 2)
  64.         cv2.imshow('src', img2)
  65.  
  66.     # -------------------------双击 结束选取-----------------------------
  67.     if event == cv2.EVENT_LBUTTONDBLCLK:
  68.         # -----------绘制感兴趣区域-----------
  69.         ROI_byMouse()
  70.         ROI_bymouse_flag = 1
  71.         lsPointsChoose = []
  72.  
  73.  
  74. def ROI_byMouse():
  75.     global src, ROI, ROI_flag, mask2
  76.     mask = np.zeros(img.shape, np.uint8) # (450, 800, 3)
  77.     pts = np.array([lsPointsChoose], np.int32)  # pts是多边形的顶点列表（顶点集）
  78.     pts = pts.reshape((-1, 1, 2))
  79.  
  80.     print(pts) #所勾选的坐标
  81.  
  82.     # 这里 reshape 的第一个参数为-1, 表明这一维的长度是根据后面的维度的计算出来的。
  83.     # OpenCV中需要先将多边形的顶点坐标变成顶点数×1×2维的矩阵，再来绘制
  84.  
  85.     # --------------画多边形---------------------
  86.     mask = cv2.polylines(mask, [pts], True, (255, 255, 255))
  87.     ##-------------填充多边形---------------------
  88.     mask2 = cv2.fillPoly(mask, [pts], (255, 255, 255))
  89.  
  90.  
  91.     cv2.imshow('mask', mask2)
  92.     cv2.imwrite('mask.jpg', mask2)
  93.  
  94.     contours, hierarchy = cv2.findContours(cv2.cvtColor(mask2, cv2.COLOR_BGR2GRAY), cv2.RETR_TREE,
  95.                                                   cv2.CHAIN_APPROX_NONE)
  96.     ROIarea = cv2.contourArea(contours[0])
  97.     print("ROIarea:", ROIarea)
  98.     ROI = cv2.bitwise_and(mask2, img)
  99.     # cv2.imwrite('ROI.jpg', ROI)
  100.     cv2.imshow('ROI', ROI)
  101.  
  102.     #暂时 下面 只针对 3 或 4 个点
  103.     crop_change = transform4pts(ROI, pts, ROI.shape) if len(pts) > 3 else transform3pts(ROI, pts, ROI.shape)
  104.     cv2.imshow('ROI_trans', crop_change)
  105.  
  106.  
  107. def ROI_test(img_path):
  108.  
  109.     img = cv2.imread(img_path)
  110.  
  111.     # 自己定义坐标点
  112.     pts = np.array(
  113.             [[[359,96]],
  114.              [[739,97]],
  115.              [[1164,661]],
  116.              [[83,663]]
  117.          ])
  118.  
  119.     crop_img = cropFill(img, pts)
  120.     cv2.imshow('crop_img', crop_img)
  121.  
  122.     crop_change = transform4pts(crop_img, pts, crop_img.shape) if len(pts) > 3 else transform3pts(crop_img, pts, crop_img.shape)
  123.     cv2.imshow('trans_img', crop_change)
  124.  
  125.     cv2.waitKey(0)
  126.     cv2.destroyAllWindows()
  127.  
  128.  
  129. # 截取并填充
  130. def cropFill(src_img, pts):
  131.  
  132.     mask = np.zeros(src_img.shape, np.uint8) # 空的画板
  133.  
  134.     # --------------画多边形---------------------
  135.     mask = cv2.polylines(mask, [pts], True, (255, 255, 255))
  136.     ##-------------填充多边形---------------------
  137.     mask2 = cv2.fillPoly(mask, [pts], (255, 255, 255))
  138.  
  139.     contours, hierarchy = cv2.findContours(cv2.cvtColor(mask2, cv2.COLOR_BGR2GRAY), cv2.RETR_TREE,
  140.                                                   cv2.CHAIN_APPROX_NONE)
  141.  
  142.     ROI = cv2.bitwise_and(mask2, src_img)
  143.  
  144.     return ROI
  145.  
  146. # 3点 透视变换
  147. def transform3pts(src_img, pts, out_size):
  148.     # print(out_size) #(730, 1176, 3) # H W
  149.     pts = pts.astype(np.float32)
  150.  
  151.     # 自定义吧
  152.     dpts = np.array([
  153.         [0, 0],
  154.         [out_size[1], 0],
  155.         [out_size[1], out_size[0]],
  156.     ], dtype=np.float32)
  157.  
  158.  
  159.     M = cv2.getAffineTransform(pts, dpts)
  160.     img_result = cv2.warpAffine(src_img, M , (out_size[1],out_size[0])) #透视变换
  161.  
  162.     return img_result
  163.  
  164. # 4点 透视变换
  165. def transform4pts(src_img, pts, out_size):
  166.     # print(out_size) #(730, 1176, 3) # H W
  167.     pts = pts.astype(np.float32)
  168.  
  169.     # 自定义吧
  170.     dpts = np.array([
  171.         [0, 0],
  172.         [out_size[1], 0],
  173.         [out_size[1], out_size[0]],
  174.         [0, out_size[0]],
  175.     ], dtype=np.float32)
  176.  
  177.     M = cv2.getPerspectiveTransform(pts, dpts) #变换矩阵  ABC变换到A'B'C'
  178.     img_result = cv2.warpPerspective(src_img, M , (out_size[1],out_size[0])) #透视变换
  179.  
  180.     return img_result
  181.  
  182. if __name__ == '__main__':
  183.  
  184.     img = cv2.imread('6.jpg') #1176,730
  185.     # ---------------------------------------------------------
  186.     # --图像预处理，设置其大小
  187.     # height, width = img.shape[:2]
  188.     # size = (int(width * 0.3), int(height * 0.3))
  189.     # img = cv2.resize(img, size, interpolation=cv2.INTER_AREA)
  190.     # ------------------------------------------------------------
  191.     ROI = img.copy()
  192.     cv2.namedWindow('src')
  193.     cv2.setMouseCallback('src', on_mouse)
  194.     cv2.imshow('src', img)
  195.     cv2.waitKey(0)
  196.     cv2.destroyAllWindows()
  197.  
  198.     # ROI_test('6.jpg')

 

四、效果

 

五、调整工具

  1. import argparse
  2. import os
  3. import tkinter as tk
  4. from tkinter import filedialog
  5.  
  6. import cv2
  7. import numpy as np
  8.  
  9. parser = argparse.ArgumentParser()
  10. # parser.add_argument('source', type=str)
  11. parser.add_argument('--height', type=int, default=1080)
  12. parser.add_argument('--width', type=int, default=1920)
  13. args = parser.parse_args()
  14. print(args)
  15.  
  16.  
  17. root = tk.Tk()
  18. root.title('不规则四边形裁剪')
  19.  
  20. pos = {
  21.     'TL': (tk.IntVar(root, value=0), tk.IntVar(root, value=0)),
  22.     'TR': (tk.IntVar(root, value=0), tk.IntVar(root, value=1600)),
  23.     'BL': (tk.IntVar(root, value=900), tk.IntVar(root, value=0)),
  24.     'BR': (tk.IntVar(root, value=900), tk.IntVar(root, value=1600)),
  25. }
  26. output = (tk.IntVar(root, value=args.height), tk.IntVar(root, value=args.width))
  27.  
  28. img = None
  29. fileName = None
  30. fileExtension = None
  31. inputRatio = None
  32.  
  33.  
  34. def mark_line(img):
  35.     if img is None:
  36.         return
  37.     img_temp = img.copy()
  38.     cv2.line(
  39.         img_temp,
  40.         (pos['TL'][1].get(), pos['TL'][0].get()),
  41.         (pos['TR'][1].get(), pos['TR'][0].get()),
  42.         (0, 0, 255), 2, cv2.LINE_AA
  43.     )
  44.     cv2.line(
  45.         img_temp,
  46.         (pos['BL'][1].get(), pos['BL'][0].get()),
  47.         (pos['BR'][1].get(), pos['BR'][0].get()),
  48.         (0, 0, 255), 2, cv2.LINE_AA
  49.     )
  50.     cv2.line(
  51.         img_temp,
  52.         (pos['TL'][1].get(), pos['TL'][0].get()),
  53.         (pos['BL'][1].get(), pos['BL'][0].get()),
  54.         (0, 0, 255), 2, cv2.LINE_AA
  55.     )
  56.     cv2.line(
  57.         img_temp,
  58.         (pos['TR'][1].get(), pos['TR'][0].get()),
  59.         (pos['BR'][1].get(), pos['BR'][0].get()),
  60.         (0, 0, 255), 2, cv2.LINE_AA
  61.     )
  62.     return img_temp
  63.  
  64.  
  65. def crop_image():
  66.     sourcePoints = np.array([
  67.         (pos['TL'][1].get(), pos['TL'][0].get()),
  68.         (pos['TR'][1].get(), pos['TR'][0].get()),
  69.         (pos['BR'][1].get(), pos['BR'][0].get()),
  70.         (pos['BL'][1].get(), pos['BL'][0].get()),
  71.     ], dtype=np.float32)
  72.     print(sourcePoints)
  73.     dstPoints = np.array([
  74.         [0, 0],
  75.         [output[1].get(), 0],
  76.         [output[1].get(), output[0].get()],
  77.         [0, output[0].get()],
  78.     ], dtype=np.float32)
  79.     print('output',output)
  80.     print(dstPoints)
  81.  
  82.     M = cv2.getPerspectiveTransform(sourcePoints, dstPoints)
  83.     print('M:',M)
  84.     print('zuobiao:',(output[1].get(), output[0].get()) )
  85.     img_result = cv2.warpPerspective(img, M, (output[1].get(), output[0].get()))
  86.     return img_result
  87.  
  88.  
  89. def show_images():
  90.     if img is None:
  91.         return
  92.  
  93.     resize_img_to_show('input', mark_line(img))
  94.     resize_img_to_show('output', crop_image(), small=True)
  95.     cv2.setMouseCallback('input', mouseClicked)
  96.  
  97.  
  98. def resize_img_to_show(name, img_temp, small=False):
  99.     if img_temp is None:
  100.         return
  101.     height = img_temp.shape[0]
  102.     width = img_temp.shape[1]
  103.     if small and height <= 700:
  104.         ratio = 1
  105.     else:
  106.         ratio = 700 / height
  107.     height = int(height * ratio)
  108.     width = int(width * ratio)
  109.     cv2.imshow(name, cv2.resize(img_temp, (width, height)))
  110.  
  111.     if name == 'input':
  112.         global inputRatio
  113.         inputRatio = ratio
  114.  
  115.  
  116. def openFile():
  117.     global img, fileName, fileExtension
  118.     file_path = filedialog.askopenfilename()
  119.     print(file_path)
  120.     if file_path is not None:
  121.         fileName = os.path.basename(file_path)
  122.         fileExtension = os.path.splitext(file_path)[1]
  123.  
  124.         img = cv2.imread(file_path, cv2.IMREAD_UNCHANGED)
  125.         pos['TL'][0].set(0)
  126.         pos['TL'][1].set(0)
  127.         pos['TR'][0].set(0)
  128.         pos['TR'][1].set(img.shape[1] - 1)
  129.         pos['BL'][0].set(img.shape[0] - 1)
  130.         pos['BL'][1].set(0)
  131.         pos['BR'][0].set(img.shape[0] - 1)
  132.         pos['BR'][1].set(img.shape[1] - 1)
  133.         show_images()
  134.  
  135.  
  136. def saveFile():
  137.     global img
  138.     if img is None:
  139.         return
  140.     out_path = filedialog.asksaveasfilename(
  141.         filetypes=[
  142.             ('PNG File', '*.png'),
  143.             ('JPEG File', '*.jpg'),
  144.             ('All files', '*'),
  145.         ],
  146.         initialfile=fileName,
  147.         defaultextension=fileExtension,
  148.     )
  149.     print(out_path)
  150.     if out_path:
  151.         img_result = crop_image()
  152.         cv2.imwrite(out_path, img_result)
  153.  
  154.  
  155. def changePos(corner, xy, diff):
  156.     def wrapper():
  157.         pos[corner][xy].set(pos[corner][xy].get() + diff)
  158.         show_images()
  159.     return wrapper
  160.  
  161.  
  162. def changeOutput(xy, diff):
  163.     def wrapper():
  164.         output[xy].set(output[xy].get() + diff)
  165.         show_images()
  166.     return wrapper
  167.  
  168.  
  169. def mouseClicked(event, p1, p0, flags, param):
  170.     if event == cv2.EVENT_LBUTTONDOWN or (event == cv2.EVENT_MOUSEMOVE and (flags & cv2.EVENT_FLAG_LBUTTON)):
  171.         p0 = int(p0 / inputRatio)
  172.         p1 = int(p1 / inputRatio)
  173.  
  174.         print(p0, p1)
  175.         minDis = 1e9
  176.         minCorner = None
  177.         for corner in pos:
  178.             tempDis = (pos[corner][0].get() - p0)**2 + (pos[corner][1].get() - p1)**2
  179.             if tempDis < minDis:
  180.                 minDis = tempDis
  181.                 minCorner = corner
  182.  
  183.         pos[minCorner][0].set(p0)
  184.         pos[minCorner][1].set(p1)
  185.         show_images()
  186.  
  187.  
  188. ROW = 0
  189. tk.Button(root, text='Open File', command=openFile).grid(row=ROW, column=0, columnspan=2, sticky='W')
  190.  
  191. for xy in [0, 1]:
  192.     ROW += 1
  193.     COL = -1
  194.     for corner in ['TL', 'TR']:
  195.         COL += 1
  196.         tk.Label(root, text='{}{}'.format(corner, xy)).grid(row=ROW, column=COL)
  197.         COL += 1
  198.         tk.Entry(root, textvariable=pos[corner][xy]).grid(row=ROW, column=COL)
  199.         for diff in [1, 10, 100]:
  200.             COL += 1
  201.             tk.Button(root, text=' {:+d} '.format(-diff), command=changePos(corner, xy, -diff)).grid(row=ROW, column=COL)
  202.             COL += 1
  203.             tk.Button(root, text=' {:+d} '.format(+diff), command=changePos(corner, xy, +diff)).grid(row=ROW, column=COL)
  204.  
  205. ROW += 1
  206. tk.Label(root, text='').grid(row=ROW, column=0)
  207.  
  208. for xy in [0, 1]:
  209.     ROW += 1
  210.     COL = -1
  211.     for corner in ['BL', 'BR']:
  212.         COL += 1
  213.         tk.Label(root, text='{}{}'.format(corner, xy)).grid(row=ROW, column=COL)
  214.         COL += 1
  215.         tk.Entry(root, textvariable=pos[corner][xy]).grid(row=ROW, column=COL)
  216.         for diff in [1, 10, 100]:
  217.             COL += 1
  218.             tk.Button(root, text=' {:+d} '.format(-diff), command=changePos(corner, xy, -diff)).grid(row=ROW, column=COL)
  219.             COL += 1
  220.             tk.Button(root, text=' {:+d} '.format(+diff), command=changePos(corner, xy, +diff)).grid(row=ROW, column=COL)
  221.  
  222. ROW += 1
  223. tk.Label(root, text='').grid(row=ROW, column=0)
  224.  
  225. ROW += 1
  226. COL = -1
  227. for xy in [0, 1]:
  228.     COL += 1
  229.     tk.Label(root, text='Out{}'.format(xy)).grid(row=ROW, column=COL)
  230.     COL += 1
  231.     tk.Entry(root, textvariable=output[xy]).grid(row=ROW, column=COL)
  232.     for diff in [1, 10, 100]:
  233.         COL += 1
  234.         tk.Button(root, text=' {:+d} '.format(-diff), command=changeOutput(xy, -diff)).grid(row=ROW, column=COL)
  235.         COL += 1
  236.         tk.Button(root, text=' {:+d} '.format(+diff), command=changeOutput(xy, +diff)).grid(row=ROW, column=COL)
  237.  
  238. ROW += 1
  239. tk.Button(root, text='Save File', command=saveFile).grid(row=ROW, column=0, columnspan=2, sticky='W')
  240.  
  241. root.mainloop()

 

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/2023面试高手/article/detail/678666
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号