導(dǎo)讀

本文主要介紹如何用OpenCV將不同的輪廓合并成一個(gè)輪廓的實(shí)現(xiàn)方法和代碼演示。

背景介紹

在圖像處理的應(yīng)用場景中常常會(huì)遇到一種情況，本來是一個(gè)整體的目標(biāo)，因?yàn)椴煌牧炼然蚱渌驅(qū)е滤指畛啥鄠€(gè)部分，這種情況在用OpenCV處理的時(shí)候會(huì)被當(dāng)成多個(gè)輪廓(如下圖所示)，那么遇到這種情況，我們?nèi)绾伟巡煌妮喞喜⒊梢粋€(gè)輪廓，然后做后續(xù)的處理呢？

實(shí)現(xiàn)方法與步驟

這里我們不用上面的繪畫圖，而是使用下面這張圖做演示：

我們的目的：將上圖中的文字輪廓看成一個(gè)整體，然后求其最小外接矩形，獲得角度，將文字旋轉(zhuǎn)水平，后續(xù)可以做簡單的文字識(shí)別。

【1】先提取文字部分輪廓(S通道閾值處理)

hsvImg = cv2.cvtColor(src,cv2.COLOR_BGR2HSV)
H,S,V = cv2.split(hsvImg)
ret, thresImg= cv2.threshold(S, 138, 255, cv2.THRESH_BINARY)

hsvImg = cv2.cvtColor(src,cv2.COLOR_BGR2HSV) H,S,V = cv2.split(hsvImg) ret, thresImg= cv2.threshold(S, 138, 255, cv2.THRESH_BINARY)

【2】中值濾波去除小雜訊

blurImg = cv2.medianBlur(thresImg,5)
cv2.imshow('blur', blurImg)

【3】查找輪廓計(jì)算輪廓最小外接矩形

contours,hierarchy = cv2.findContours(blurImg, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
merge_list = []
for cnt in contours:
  rect = cv2.minAreaRect(cnt)
 
  box = cv2.boxPoints(rect)
  box = np.int0(box)
  split_res = cv2.drawContours(split_res,[box],0,(0,0,255),2)
  merge_list.append(cnt)

hsvImg = cv2.cvtColor(src,cv2.COLOR_BGR2HSV) H,S,V = cv2.split(hsvImg) ret, thresImg= cv2.threshold(S, 138, 255, cv2.THRESH_BINARY)

【4】輪廓合并成一個(gè)繪制最小外接矩形

contours_merge = np.vstack([merge_list[0],merge_list[1]])
for i in range(2, len(merge_list)):
  contours_merge = np.vstack([contours_merge,merge_list[i]])

rect2 = cv2.minAreaRect(contours_merge)
box2 = cv2.boxPoints(rect2)
box2 = np.int0(box2)
merge_res = cv2.drawContours(merge_res,[box2],0,(0,255,0),2)

完整代碼與效果：

import numpy as np
import cv2

src = cv2.imread('A.jpg')
cv2.imshow('src', src)

split_res = src.copy()#顯示每個(gè)輪廓結(jié)構(gòu)
merge_res = src.copy()#顯示合并后輪廓結(jié)構(gòu)

# 記錄開始時(shí)間
start = cv2.getTickCount()
hsvImg = cv2.cvtColor(src,cv2.COLOR_BGR2HSV)
H,S,V = cv2.split(hsvImg)
ret, thresImg= cv2.threshold(S, 138, 255, cv2.THRESH_BINARY)
cv2.imshow('threshold', thresImg)
blurImg = cv2.medianBlur(thresImg,5)
cv2.imshow('blur', blurImg)
 
contours,hierarchy = cv2.findContours(blurImg, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

merge_list = []
for cnt in contours:
  rect = cv2.minAreaRect(cnt)
 
  box = cv2.boxPoints(rect)
  box = np.int0(box)
  split_res = cv2.drawContours(split_res,[box],0,(0,0,255),2)
  merge_list.append(cnt)
cv2.imshow('split_res', split_res)
cv2.imwrite('split_res.jpg', split_res)

contours_merge = np.vstack([merge_list[0],merge_list[1]])
for i in range(2, len(merge_list)):
  contours_merge = np.vstack([contours_merge,merge_list[i]])

rect2 = cv2.minAreaRect(contours_merge)
box2 = cv2.boxPoints(rect2)
box2 = np.int0(box2)
merge_res = cv2.drawContours(merge_res,[box2],0,(0,255,0),2)
cv2.imshow('merge_res', merge_res)
cv2.imwrite('merge_res.jpg', merge_res)

# 記錄結(jié)束時(shí)間   
end = cv2.getTickCount()
# 運(yùn)行耗時(shí)
use_time = (end - start) / cv2.getTickFrequency()
print('use-time: %.3fs' % use_time)

cv2.waitKey(0)
cv2.destroyAllWindows()
print ('finish')