利用Python將圖片中扭曲矩形的復原

更新時間：2020年09月07日 11:00:39 作者：颹蕭蕭

這篇文章主要給大家介紹了關(guān)于如何利用Python將圖片中扭曲矩形復原的相關(guān)資料，文中通過示例代碼介紹的非常詳細，對大家的學習或者工作具有一定的參考學習價值，需要的朋友們下面隨著小編來一起學習學習吧

前言

日常生活中，手殘黨們經(jīng)常會把一些照片拍歪，比如拍個證件、試卷、PPT什么的，

比如下面這本書的封面原本是個矩形，隨手一拍就成了不規(guī)則四邊形

想要把它變成規(guī)則的矩形，有什么辦法呢？

你一定想到了 PS，因為網(wǎng)上很多教程就是這么教的

打開PS ⟹\Longrightarrow⟹ 濾鏡 ⟹\Longrightarrow⟹鏡頭校正 ⟹\Longrightarrow⟹自定 ⟹\Longrightarrow⟹變換

可是這樣手工的調(diào)整數(shù)值實在是費勁喲?。?/p>

下面我們來寫個程序，自動將圖片恢復到任意矩形！

程序?qū)崿F(xiàn)

1. 打開圖片并顯示

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. 在圖片中標記四個角點

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{}個頂點: ({},{})'.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),     # 坐標
   cv.FONT_HERSHEY_PLAIN,
   5,      # 字號
   (0,0,255),    # 字體顏色
   thickness = 2   # 粗細
  )

  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. 設(shè)定輸出矩形大小

# W = int(input('輸出圖片寬度:'))
# H = int(input('輸出圖片高度:'))
W,H = 600, 800
target_points = [(0,0),(W,0),(W,H),(0,H)]

4. 求解透視變換矩陣

points, target_points = np.array(points, dtype=np.float32), np.array(target_points, dtype=np.float32)
M = cv.getPerspectiveTransform(points, target_points)
print('透視變換矩陣:',M)

5. 透視變換并保存圖片

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)

結(jié)果如下