基于图像处理和K近邻的车牌号识别

àì夳堔傛蜴生んèń 2021-06-11 15:11 504阅读 0赞

[源码下载][Link 1]  
CNN流程：  
![这里写图片描述][SouthEast]  
整个流程：  
![这里写图片描述][SouthEast 1]

generate.py：分离整张字符图像代码
    # 将字符图片分离成字符
    import os
    import cv2
    import numpy as np
    # 提取字符函数
    def extract_chars(img):
        # 使用二值化处理
        ret, bwimg = cv2.threshold(img, 64, 255, cv2.THRESH_BINARY)
        # print(bwimg)
        # 定义一个3*3的十字结构
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
        # 对图像进行腐蚀,缩小边缘
        erode = cv2.erode(bwimg, kernel, iterations = 1)
        # 图像非运算，黑白颠倒
        bw_image = cv2.bitwise_not(erode)
        # cv2.imshow("Image", img)
        # cv2.waitKey(0)
        # 检测图像轮廓，返回轮廓属性
        contours = cv2.findContours(bw_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)[1]
        # 生成轮廓框
        bounding_boxes = []
        for contour in contours:
            # 轮廓基点和长宽
            x,y,w,h = cv2.boundingRect(contour)
            x,y,w,h = x-2, y-2, w+4, h+4
            bounding_boxes.append((x,y,w,h))
        # 裁出字符图片
        characters = []
        for bbox in bounding_boxes:
            x,y,w,h = bbox
            char_image = img[y:y+h,x:x+w]
            characters.append(char_image)
        return characters
    # 存储字符函数
    def output_chars(chars, labels):
        for i, char in enumerate(chars):
            # cv2.imshow("Image", char)
            # cv2.waitKey(0)
            filename = "chars/%s.png" % labels[i]
            # 图像长宽放大三倍，邻域双三次插值
            char = cv2.resize(char
                , None
                , fx=3
                , fy=3
                , interpolation=cv2.INTER_CUBIC)
            # char = cv2.resize(char,(90,140))
            cv2.imwrite(filename, char)
    # 生成文件夹
    if not os.path.exists("chars"):
        os.makedirs("chars")
    # 读取灰度数字和字母图片
    img_digits = cv2.imread("ch/digit.jpg", 0)
    img_letters = cv2.imread("ch/letter.jpg", 0)
    # 提取字符
    digits = extract_chars(img_digits)
    letters = extract_chars(img_letters)
    DIGITS = [1, 0, 9, 8 ,7, 6, 5, 4, 3, 2]
    # LETTERS = [chr(ord('A') + i) for i in range(25,-1,-1)]
    LETTERS = ['N', 'M', 'L', 'K' ,'J', 'I', 'H', 'G', 'F', 'E', 'D', 'C', 'B', 'A', 'Z', 'Y', 'X', 'W', 'V', 'U', 'T', 'S', 'R', 'Q', 'P', 'O']
    # 以字符命名方式存储字符图片
    output_chars(digits, DIGITS)
    output_chars(letters, LETTERS) 
    
    train.py: 提取字符特征，KNN初步测试
    # 将字符图片转成特征，并生成各个特征的label
    import cv2
    import numpy as np
    # 生成车牌号中字符 ord由字符生成Ascii码 chr由Ascii码生成字符
    CHARS = [chr(ord('0') + i) for i in range(10)] + [chr(ord('A') + i) for i in range(26)]
    # print(CHARS)
    # 加载字符图片函数
    def load_char_images():
        characters = {}
        for char in CHARS:
            # 以灰度模式读取
            char_img = cv2.imread("chars/%s.png" % char, 0)
            characters[char] = char_img
        return characters
    # 加载字符图片
    characters = load_char_images()
    # 特征长度为100
    samples =  np.empty((0,100))
    for char in CHARS:
        char_img = characters[char]
        # 将图片缩放成10*10,保持特征长度的一致性
        small_char = cv2.resize(char_img,(10,10))
        # 图片进行切片成1*100，一维向量
        sample = small_char.reshape((1,100))
        # 添加进samples
        samples = np.append(samples,sample,0)
    # 生成字符label，用列存储
    label = np.array([ord(c) for c in CHARS],np.float32)
    label = label.reshape((label.size,1))
    # 存储字符特征和字符label
    np.savetxt('char_samples.data',samples)
    np.savetxt('char_label.data',label)
    # 模型训练
    # 读取字符特征和标签
    samples = np.loadtxt('char_samples.data',np.float32)
    label = np.loadtxt('char_label.data',np.float32)
    label = label.reshape((label.size,1))
    # 使用K近邻进行学习
    model = cv2.ml.KNearest_create()
    model.train(samples, cv2.ml.ROW_SAMPLE, label)
    # 读取一个测试样本
    # img = cv2.imread("chars/2.png",0)
    total=0
    for char in CHARS:
        img = cv2.imread("chars/%s.png" %char,0)
        small_img = cv2.resize(img,(10,10))
        ret, small_img = cv2.threshold(small_img, 64, 255, cv2.THRESH_BINARY)
        # print(small_img)
        small_img = small_img.reshape((1,100))
        small_img = np.float32(small_img)
        # 使用K近邻进行预测
        retval, results, neigh_resp, dists = model.findNearest(small_img, k = 1)
        print(chr(results[0][0]))
        if(chr(results[0][0])==char):
            total += 1
    print(1.0*total/len(CHARS))
    
    main.py: 车牌号图像处理并识别
    # 主程序,识别车牌号
    import cv2
    import numpy as np
    import sys
    import argparse
    import os
    def reduce_colors(img, n):
        # 图片切片成3列
        Z = img.reshape((-1,3))
        # 转成float32类型,运算快
        Z = np.float32(Z)
        # 使用k均值进行聚类
        # 迭代停止的模式选择:满足精确度+最大迭代次数
        criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0)
        # K值,分类数
        K = n
        # 重复进行多少次,将返回最好一次结果
        # 紧密度，返回每个点到相应重心的距离的平方和; labels：标志数组;centers：由聚类的中心组成的数组。
    ret,label,center=cv2.kmeans(Z,K,None,criteria,10,cv2.KMEANS_RANDOM_CENTERS)
        # 图片重新转成uint8数据类型
        center = np.uint8(center)
        res = center[label.flatten()]
        # 返回图片
        res2 = res.reshape((img.shape))
        return res2 
    # 图片处理函数
    def clean_image(img):
        # RGB转灰度图
        gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        # 缩放图片
        resized_img = cv2.resize(gray_img
            , None
            , fx=4.0
            , fy=4.0
            , interpolation=cv2.INTER_CUBIC)
        # 进行高斯滤波
        resized_img = cv2.GaussianBlur(resized_img,(5,5),0)
        # 存储高斯结果
        cv2.imwrite('temp/licence_plate_gauss.png', resized_img)
        # 直方图均衡化
        equalized_img = cv2.equalizeHist(resized_img)
        # 存储均衡化直方图
        cv2.imwrite('temp/licence_plate_equ.png', equalized_img)
        # 图片经过聚类处理
        reduced = cv2.cvtColor(reduce_colors(cv2.cvtColor(equalized_img, cv2.COLOR_GRAY2BGR), 8), cv2.COLOR_BGR2GRAY)
        cv2.imwrite('temp/licence_plate_red.png', reduced)
        # 限定阈值二值化图片
        ret, bwimg = cv2.threshold(reduced, 64, 255, cv2.THRESH_BINARY)
        # 存储二值图片
        cv2.imwrite('temp/licence_plate_bwimg.png', bwimg) 
        # 定义一个3*3的十字结构
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (4, 4))
        # 对图像进行腐蚀
        erode = cv2.erode(bwimg, kernel, iterations = 1)
        # 存储腐蚀图片
        cv2.imwrite('temp/licence_plate_erode.png', erode)
        return erode
    # 提取字符函数
    def extract_characters(img):
        # 非运算,将数字部分转成白颜色
        bw_image = cv2.bitwise_not(img)
        cv2.imwrite("temp/bw_image.png",bw_image)
        # 查找边缘
        contours = cv2.findContours(bw_image, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)[1]
        # print(contours)
        char_mask = np.zeros_like(img)
        bounding_boxes = []
        for contour in contours:
            x,y,w,h = cv2.boundingRect(contour)
            area = w * h
            center = (x + w/2, y + h/2)
            # 对于区域在1000~10000之间的才算字符
            if (area > 1800) and (area < 10000):
                x,y,w,h = x-4, y-4, w+8, h+8
                # 存储图片边框信息
                bounding_boxes.append((center, (x,y,w,h)))
                # 绘画矩形,使用灰度255白色,线宽-1表示填满
                cv2.rectangle(char_mask,(x,y),(x+w,y+h),255,-1)
        # 存储矩形区域
        cv2.imwrite('temp/licence_plate_char_mask.png', char_mask)
        # 将原图和提取的字符图进行与与运算提取出字符图,在进行非运算
        clean = cv2.bitwise_not(cv2.bitwise_and(char_mask, char_mask, mask = bw_image))
        # 根据字符中心进行从左到右排序
        bounding_boxes = sorted(bounding_boxes, key=lambda item: item[0][0])  
        # 裁出字符图片
        characters = []
        for center, bbox in bounding_boxes:
            x,y,w,h = bbox
            char_image = clean[y:y+h,x:x+w]
            characters.append((bbox, char_image))
        # 返回整张字符图和裁剪后字符图
        return clean, characters
    # 高亮字符图
    def highlight_characters(img, chars, file):
        # 图片转RGB
        # if(file=="CU5600"):
        output_img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
        for bbox, char_img in chars:
            x,y,w,h = bbox
            # 将所选区域用白色绘制
            cv2.rectangle(output_img,(x,y),(x+w,y+h),255,1)
        # 存储高亮字符图片
        cv2.imwrite('temp/licence_plate_out.png', output_img)
        return output_img
    # 开始执行
    #命令形式： python main.py --image T1.jpg
    # 从命令行获取要读取的车牌号图片 
    # ap = argparse.ArgumentParser()
    # ap.add_argument("-i", "--image", required = True, help = "输入图片路径")
    # args = vars(ap.parse_args())
    # # 读取图片
    # img = cv2.imread(args["image"])
    # 读取字符特征和标签
    samples = np.loadtxt('char_samples.data',np.float32)
    label = np.loadtxt('char_label.data',np.float32)
    label = label.reshape((label.size,1))
    # 使用K近邻模型并训练
    model = cv2.ml.KNearest_create()
    model.train(samples, cv2.ml.ROW_SAMPLE, label)
    # 重定向输出流
    orig_stdout = sys.stdout
    f = open('output.txt','w')
    sys.stdout = f
    path = "data" #文件夹目录 
    files= os.listdir(path)
    total=0
    for file in files:
        # 读取图片
        img = cv2.imread(path+"/"+file)
        # print(file)
        img = cv2.resize(img,(108,24))
        # 处理图片
        img = clean_image(img)
        # 提取图片中字符
        clean_img, chars = extract_characters(img)
        # 高亮图片中字符
        output_img = highlight_characters(clean_img, chars,file[0:-4])
        plate_chars = ""
        for bbox, char_img in chars:
            # 将提取的char转成字符特征形式
            # print("char_img.shape")
            # print(char_img.shape)
            small_img = cv2.resize(char_img,(10,10))
            small_img = small_img.reshape((1,100))
            small_img = np.float32(small_img)
            retval, results, neigh_resp, dists = model.findNearest(small_img, k = 1)
            plate_chars += str(chr((results[0][0])))
        # 存储识别的车牌号
        print("识别车牌号: %s" % plate_chars)
        print("实际车牌号: %s" % file[0:-4])
        print()
        if(plate_chars==file[0:-4]):
            total = total + 1
    sys.stdout = orig_stdout
    f.close()
    print("准确率",1.00*total/len(files))

[Link 1]: http://download.csdn.net/download/qq_16234613/10208835
[SouthEast]: /images/20210520/8ad57f4605404d928021bcb6d5dc8ad1.png
[SouthEast 1]: /images/20210520/42dda31b88ae443f9ab60b53fb7ddff1.png