import cv2
import numpy as np
def lpr(filename):
  img = cv2.imread(filename)
  # 預(yù)處理，包括灰度處理，高斯濾波平滑處理，Sobel提取邊界，圖像二值化
  # 對(duì)于高斯濾波函數(shù)的參數(shù)設(shè)置，第四個(gè)參數(shù)設(shè)為零，表示不計(jì)算y方向的梯度，原因是車牌上的數(shù)字在豎方向較長(zhǎng)，重點(diǎn)在于得到豎方向的邊界
  # 對(duì)于二值化函數(shù)的參數(shù)設(shè)置，第二個(gè)參數(shù)設(shè)為127，是二值化的閾值，是一個(gè)經(jīng)驗(yàn)值
  gray_img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
  GaussianBlur_img = cv2.GaussianBlur(gray_img, (3, 3), 0)
  Sobel_img = cv2.Sobel(GaussianBlur_img, -1, 1, 0, ksize=3)
  ret, binary_img = cv2.threshold(Sobel_img, 127, 255, cv2.THRESH_BINARY)
  # 形態(tài)學(xué)運(yùn)算
  kernel = np.ones((5, 15), np.uint8)
  # 先閉運(yùn)算將車牌數(shù)字部分連接，再開運(yùn)算將不是塊狀的或是較小的部分去掉
  close_img = cv2.morphologyEx(binary_img, cv2.MORPH_CLOSE, kernel)
  open_img = cv2.morphologyEx(close_img, cv2.MORPH_OPEN, kernel)
  # kernel2 = np.ones((10, 10), np.uint8)
  # open_img2 = cv2.morphologyEx(open_img, cv2.MORPH_OPEN, kernel2)
  # 由于部分圖像得到的輪廓邊緣不整齊，因此再進(jìn)行一次膨脹操作
  element = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
  dilation_img = cv2.dilate(open_img, element, iterations=3)
  # 獲取輪廓
  contours, hierarchy = cv2.findContours(dilation_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  # 測(cè)試邊框識(shí)別結(jié)果
  # cv2.drawContours(img, contours, -1, (0, 0, 255), 3)
  # cv2.imshow("lpr", img)
  # cv2.waitKey(0)
  # 將輪廓規(guī)整為長(zhǎng)方形
  rectangles = []
  for c in contours:
    x = []
    y = []
    for point in c:
      y.append(point[0][0])
      x.append(point[0][1])
    r = [min(y), min(x), max(y), max(x)]
    rectangles.append(r)
  # 用顏色識(shí)別出車牌區(qū)域
  # 需要注意的是這里設(shè)置顏色識(shí)別下限low時(shí)，可根據(jù)識(shí)別結(jié)果自行調(diào)整
  dist_r = []
  max_mean = 0
  for r in rectangles:
    block = img[r[1]:r[3], r[0]:r[2]]
    hsv = cv2.cvtColor(block, cv2.COLOR_BGR2HSV)
    low = np.array([100, 60, 60])
    up = np.array([140, 255, 255])
    result = cv2.inRange(hsv, low, up)
    # 用計(jì)算均值的方式找藍(lán)色最多的區(qū)塊
    mean = cv2.mean(result)
    if mean[0] > max_mean:
      max_mean = mean[0]
      dist_r = r
  # 畫出識(shí)別結(jié)果，由于之前多做了一次膨脹操作，導(dǎo)致矩形框稍大了一些，因此這里對(duì)于框架+3-3可以使框架更貼合車牌
  cv2.rectangle(img, (dist_r[0]+3, dist_r[1]), (dist_r[2]-3, dist_r[3]), (0, 255, 0), 2)
  cv2.imshow("lpr", img)
  cv2.waitKey(0)
# 主程序
for i in range(5):
  lpr(str(i+1) + ".jpg")