代码收藏家技术教程 2022-07-18

运用opencv 进行车牌的识别（1）

文章目录

简介：

项目准备

项目阶段

车牌的处理和提取

原理：

实现：

筛选图像

分隔字符

函数入口定义

效果展示

完整代码

简介：

**无人停车场可以说是目前人工智能应用最普通的一个，核心的技术在于车牌的识别，那么作为人工智能的狂热者，自然不能错过这么好的项目。赶紧开始吧！ **

项目准备

python 环境
opencv环境
numpy环境

项目阶段

提取车牌部分
进行车牌的字符分隔
进行车牌的字符识别
进行车牌的倾斜调整等特殊情况优化
项目的落地和部署

车牌的处理和提取

原理：

对车牌进行去噪，运用opencv的形态学操作等方法分割成多个部分

根据车牌的比例，进行轮廓的筛选，识别出车牌部分

实现：

读取图片，灰度处理

img = cv.imread(path)
img_gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

进行高斯处理，运用sobel算子计算轮廓

   img_gs = cv.GaussianBlur(img_gray, (5, 5), 0, 0, cv.BORDER_DEFAULT)
   #sobel
   sobel_x = cv.Sobel(img_gs, cv.CV_16S, 1, 0)
   # sobel_y = cv.Sobel(img_gs, cv.CV_16S, 0, 1)
   absX = cv.convertScaleAbs(sobel_x)
   # absY = cv.convertScaleAbs(sobel_y)
   # img = cv.addWeighted(sobel_x, 0.5, sobel_y, 0.5, 0)
   image = absX

二值化处理，自适应

ret, image = cv.threshold(image, 0, 255, cv.THRESH_OTSU)

进行一次闭合操作, 连接一些小白点

kernelX = cv.getStructuringElement(cv.MORPH_RECT, (17, 5))
image = cv.morphologyEx(image, cv.MORPH_CLOSE, kernelX, iterations=3)

进行腐蚀膨胀操作，去除一些噪点，扩大轮廓区域

#去除小白点
    kernelX = cv.getStructuringElement(cv.MORPH_RECT, (20, 1))
    kernelY = cv.getStructuringElement(cv.MORPH_RECT, (1, 19))
    image = cv.erode(image, kernelX)
    image = cv.erode(image, kernelY)
    image = cv.dilate(image, kernelX)
    image = cv.dilate(image, kernelY)

中值滤波，为了降低噪点

 # 中值滤波
    image = cv.medianBlur(image, 15)

进行轮廓检测

# 轮廓检测
    contours, hierarchy = cv.findContours(image, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)

筛选图像

一般车牌比例为长：宽 = 1:3~1:6, 根据这个原理

def filter_photo(img, contours):
    # img_copy = img.copy()
    for cont in contours:
        x, y, w, h = cv.boundingRect(cont)
        # cv.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 3)
        area = cv.contourArea(cont)
        w_rate_h = float(w/h)
        if(w_rate_h >= 3 and w_rate_h <= 6):
            lpimage = img[y: y+h, x: x+w]
            break
    
    return lpimage, x, y, w, h

分隔字符

字符的分割比较简单，将图片进行二值化，提取白色的数字即可

# 字符分割
def split_img(img1):
    img = cv.cvtColor(img1, cv.COLOR_BGR2GRAY)
    ret, before_threshold = cv.threshold(img, 0, 255, cv.THRESH_OTSU)
    img = cv.GaussianBlur(img, (3, 3), 0)
    ret, threshold = cv.threshold(img, 0, 255, cv.THRESH_OTSU)
    # 变胖
    kernel = cv.getStructuringElement(cv.MORPH_RECT, (3, 3))
    edge1 = cv.morphologyEx(threshold, cv.MORPH_CLOSE, kernel)
    # 中值去噪
    edge2 = cv.medianBlur(edge1, 3)
    # 提取轮廓
    contours, hierarchy = cv.findContours(edge2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    words = []
	# 这里需要对选区进行处理，过滤高和宽的比在0.5~2.3内的数字
    for i, cont in enumerate(contours):
        word = []
        x, y, w, h = cv.boundingRect(cont)
        ratew_h = h/w
        # cv.rectangle(img1, (x, y), (x+w, y+h), (0, 255, 0), 3)
        if ratew_h > 0.5 and ratew_h < 2.3:
            word.append((x, y, w, h))
            words.append(word)

            
    words = sorted(words, key=lambda s: s[0], reverse=False)
    # cv.drawContours(img1, contours, -1, (0, 255, 0), 1)
    # 截图处理
    for i, area in enumerate(words):
        (x, y, w, h) = area[0]
        num_img = before_threshold[y: y+ h, x: x+ w]
        # num_img = cv.cvtColor(num_img, cv.COLOR_BGR2GRAY)
        # ret, num_img = cv.threshold(num_img, 177, 255, cv.THRESH_BINARY)
        cv.imwrite('./split/i%s.png'%i, num_img)

函数入口定义

def main():
    path = './car1.png'
    img, contours = hander_img(path)
    area, x, y, w, h = filter_photo(img, contours)
    split_img(area)
    # cv.imwrite('./pai.png', area)
    # cv.imshow('img2', area)
    # cv.waitKey(0)
    # cv.destroyAllWindows()
main()

效果展示

车牌

分割结果

准备工作完毕，接下来只需要进行字符的识别工作就可以了

完整代码

车牌提取

import cv2 as cv
import numpy as np
def hander_img(path):
    img = cv.imread(path)
    img_gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

    img_gs = cv.GaussianBlur(img_gray, (5, 5), 0, 0, cv.BORDER_DEFAULT)
    #sobel
    sobel_x = cv.Sobel(img_gs, cv.CV_16S, 1, 0)
    # sobel_y = cv.Sobel(img_gs, cv.CV_16S, 0, 1)
    absX = cv.convertScaleAbs(sobel_x)
    # absY = cv.convertScaleAbs(sobel_y)
    # img = cv.addWeighted(sobel_x, 0.5, sobel_y, 0.5, 0)
    image = absX
    ret, image = cv.threshold(image, 0, 255, cv.THRESH_OTSU)
    kernelX = cv.getStructuringElement(cv.MORPH_RECT, (17, 5))
    image = cv.morphologyEx(image, cv.MORPH_CLOSE, kernelX, iterations=3)
    #去除小白点
    kernelX = cv.getStructuringElement(cv.MORPH_RECT, (20, 1))
    kernelY = cv.getStructuringElement(cv.MORPH_RECT, (1, 19))
    image = cv.erode(image, kernelX)
    image = cv.erode(image, kernelY)
    image = cv.dilate(image, kernelX)
    image = cv.dilate(image, kernelY)
    # 中值滤波
    image = cv.medianBlur(image, 15)
    # 轮廓检测
    contours, hierarchy = cv.findContours(image, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    
    # cv.drawContours(img, contours, -1, (0, 255, 0),3)
    # cv.imshow('img', image)
    # cv.imshow('img2', image)
    cv.waitKey(0)
    cv.destroyAllWindows()
    return img, contours

车牌定位

def filter_photo(img, contours):
    # img_copy = img.copy()
    for cont in contours:
        x, y, w, h = cv.boundingRect(cont)
        # cv.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 3)
        area = cv.contourArea(cont)
        w_rate_h = float(w/h)
        if(w_rate_h >= 3 and w_rate_h <= 6):
            lpimage = img[y: y+h, x: x+w]
            break
    
    return lpimage, x, y, w, h
    cv.imshow("photo", img)
    cv.waitKey(0)
    cv.destroyAllWindows()

分割字符

# 字符分割
def split_img(img1):
    img = cv.cvtColor(img1, cv.COLOR_BGR2GRAY)
    ret, before_threshold = cv.threshold(img, 0, 255, cv.THRESH_OTSU)
    img = cv.GaussianBlur(img, (3, 3), 0)
    ret, threshold = cv.threshold(img, 0, 255, cv.THRESH_OTSU)
    # 变胖
    kernel = cv.getStructuringElement(cv.MORPH_RECT, (3, 3))
    edge1 = cv.morphologyEx(threshold, cv.MORPH_CLOSE, kernel)
    # 中值去噪
    edge2 = cv.medianBlur(edge1, 3)
    # 提取轮廓
    contours, hierarchy = cv.findContours(edge2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    words = []

    for i, cont in enumerate(contours):
        word = []
        x, y, w, h = cv.boundingRect(cont)
        ratew_h = h/w
        # cv.rectangle(img1, (x, y), (x+w, y+h), (0, 255, 0), 3)
        if ratew_h > 0.5 and ratew_h < 2.3:
            word.append((x, y, w, h))
            words.append(word)

            
    words = sorted(words, key=lambda s: s[0], reverse=False)
    # cv.drawContours(img1, contours, -1, (0, 255, 0), 1)
    # 截图处理
    for i, area in enumerate(words):
        (x, y, w, h) = area[0]
        num_img = before_threshold[y: y+ h, x: x+ w]
        # num_img = cv.cvtColor(num_img, cv.COLOR_BGR2GRAY)
        # ret, num_img = cv.threshold(num_img, 177, 255, cv.THRESH_BINARY)
        cv.imwrite('./split/i%s.png'%i, num_img)

main

def main():
    path = './car2.png'
    img, contours = hander_img(path)
    area, x, y, w, h = filter_photo(img, contours)
    split_img(area)
    # cv.imwrite('./pai.png', area)
    cv.imshow('img2', area)
    cv.waitKey(0)
    cv.destroyAllWindows()
main()

来源：MioeC