利用Python将图片中扭曲矩形的复原

前言

日常生活中，手残党们经常会把一些照片拍歪，比如拍个证件、试卷、PPT什么的，

比如下面这本书的封面原本是个矩形，随手一拍就成了不规则四边形

想要把它变成规则的矩形，有什么办法呢？

你一定想到了 PS，因为网上很多教程就是这么教的

打开PS ⟹\\Longrightarrow⟹ 滤镜 ⟹\\Longrightarrow⟹镜头校正 ⟹\\Longrightarrow⟹自定 ⟹\\Longrightarrow⟹变换

可是这样手工的调整数值实在是费劲哟！！

下面我们来写个程序，自动将图片恢复到任意矩形！

程序实现

1. 打开图片并显示

import cv2 as cv
import numpy as np

path = \'./book.jpg\'
src = cv.imread(path)

def get_window_size(src, bound=600):
 h,w = src.shape[0], src.shape[1]
 if h > w:
  h, w = bound, int(w*bound/h)
 else:
  h, w = int(h*bound/w), bound
 return (h, w)

h, w = get_window_size(src)

win_name = \'RecAdjust\'
cv.namedWindow(win_name, cv.WINDOW_NORMAL)
cv.resizeWindow(win_name, width=w, height=h)
cv.imshow(win_name, src)
cv.waitKey(0)
cv.destroyAllWindows()

2. 在图片中标记四个角点

src_copy = src.copy()

class Indexer:
 def __init__(self, bound=4):
  self.id = 0
  self.bound=bound
 
 def get_id(self):
  self.id = (self.id+1)%self.bound
  return (self.id-1)%self.bound+1

def on_EVENT_LBUTTONDOWN(event, x, y, flags, param):
 if event == cv.EVENT_LBUTTONDOWN:
  
  img = param[\'src\']
  win_name = param[\'window\']
  indexer = param[\'indexer\']
  points = param[\'points\']
  
  curr_id = indexer.get_id()
  points.append((x,y))
  print(\'第{}个顶点: ({},{})\'.format(curr_id, x, y))
  
  cv.circle(img, (x, y), 10, (0, 0, 255), thickness = 2)
  cv.putText(
   img,  
   str(curr_id),   # 文字
   (x, y),     # 坐标
   cv.FONT_HERSHEY_PLAIN,
   5,      # 字号
   (0,0,255),    # 字体颜色
   thickness = 2   # 粗细
  )

  cv.imshow(win_name, img)

points = []
indexer = Indexer()
win_name = \'RecAdjust\'
cv.namedWindow(win_name, cv.WINDOW_NORMAL)
cv.resizeWindow(win_name, width=w, height=h)
cv.imshow(win_name, src)
cv.setMouseCallback(win_name, on_EVENT_LBUTTONDOWN, param={\'src\':src,\'window\':win_name, \'indexer\':indexer, \'points\':points})
cv.waitKey(0)
cv.destroyAllWindows()
print(points)

3. 设定输出矩形大小

# W = int(input(\'输出图片宽度:\'))
# H = int(input(\'输出图片高度:\'))
W,H = 600, 800
target_points = [(0,0),(W,0),(W,H),(0,H)]

4. 求解透视变换矩阵

points, target_points = np.array(points, dtype=np.float32), np.array(target_points, dtype=np.float32)
M = cv.getPerspectiveTransform(points, target_points)
print(\'透视变换矩阵:\',M)

5. 透视变换并保存图片

result = cv.warpPerspective(src_copy, M, (0, 0))
result = result[:H, :W]
win_name = \'Result\'
cv.namedWindow(win_name, cv.WINDOW_NORMAL)
cv.resizeWindow(win_name, width=W, height=H)
cv.imshow(win_name, result)
cv.waitKey(0)
cv.destroyAllWindows()

output_file = \'result.jpg\'
cv.imwrite(output_file, result)

结果如下