python与tensorflow实现车牌识别

python实现深度学习+传统方法的车牌识别
系统流程图：

其中，进行灰度转化，高斯滤波，形态学操作等方式，对输入图像进行预处理，处理过程图如下图所示：



处理代码为：

def pre_process(orig_img):

    gray_img = cv2.cvtColor(orig_img, cv2.COLOR_BGR2GRAY)
    # cv2.imshow('gray_img', gray_img)

    # 均值滤波
    blur_img = cv2.blur(gray_img, (3, 3))
    # cv2.imshow('blur', blur_img)

    # 边缘检测
    sobel_img = cv2.Sobel(blur_img, cv2.CV_16S, 1, 0, ksize=3)
    # 用convertScaleAbs()函数将其转回原来的uint8形式
    sobel_img = cv2.convertScaleAbs(sobel_img)
    # cv2.imshow('sobel', sobel_img)

    hsv_img = cv2.cvtColor(orig_img, cv2.COLOR_BGR2HSV)

    h, s, v = hsv_img[:, :, 0], hsv_img[:, :, 1], hsv_img[:, :, 2]
    # 黄色色调区间[26，34],蓝色色调区间:[100,124]，绿色色调区间：[35,77]
    # | ((h > 35) & (h < 77))
    choose_img = (((h > 26) & (h < 34)) | ((h > 100) & (h < 124))| ((h > 35) & (h < 77))) & (s > 70) & (v > 70)
    choose_img = choose_img.astype('float32')

    # 挑选出黄牌、蓝牌、绿牌的区域
    mix_img = np.multiply(sobel_img, choose_img)
    # cv2.imshow('mix', mix_img)

    mix_img = mix_img.astype(np.uint8)

    ret, binary_img = cv2.threshold(mix_img, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
    #cv2.imshow('binary',binary_img)

    # 函数可用于构造一个特定大小和形状的结构元素，用于图像形态学处理。
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(21,5))
    # 闭运算是对图像先膨胀再腐蚀，可以排除小型黑洞
    close_img = cv2.morphologyEx(binary_img, cv2.MORPH_CLOSE, kernel)
    # cv2.imshow('close', close_img)

    return close_img

采用传统方法与深度学习方法结合的方式，定位车牌图像，定位结果图为：

识别采用改进的Lenet卷积神经网络进行字符识别训练：训练集截图如图

训练过程主要代码如下：

# conv1、pooling1
        cw1 = tf.Variable(tf.random_normal(shape=[3, 3, 1, 32], stddev=0.01), dtype=tf.float32)
        cb1 = tf.Variable(tf.random_normal(shape=[32]), dtype=tf.float32)
        conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x_input,filter=cw1,strides=[1,1,1,1],padding='SAME'),cb1))
        conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
        conv1 = tf.nn.dropout(conv1, self.keep_place)

        # conv2、pooling2
        cw2 = tf.Variable(tf.random_normal(shape=[3, 3, 32, 64], stddev=0.01), dtype=tf.float32)
        cb2 = tf.Variable(tf.random_normal(shape=[64]), dtype=tf.float32)
        conv2 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv1,filter=cw2,strides=[1,1,1,1],padding='SAME'),cb2))
        conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
        conv2 = tf.nn.dropout(conv2, self.keep_place)

        # conv3、pooling3
        cw3 = tf.Variable(tf.random_normal(shape=[3, 3, 64, 128], stddev=0.01), dtype=tf.float32)
        cb3 = tf.Variable(tf.random_normal(shape=[128]), dtype=tf.float32)
        conv3 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv2,filter=cw3,strides=[1,1,1,1],padding='SAME'),cb3))
        conv5 = tf.nn.max_pool(conv3, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
        conv5 = tf.nn.dropout(conv5, self.keep_place)

        conv_out = tf.reshape(conv5, shape=[-1, 3 * 3 * 128])

        fw1 = tf.Variable(tf.random_normal(shape=[3 * 3 * 128, 1024], stddev=0.01), dtype=tf.float32)
        fb1 = tf.Variable(tf.random_normal(shape=[1024]), dtype=tf.float32)
        fully1 = tf.nn.relu(tf.add(tf.matmul(conv_out, fw1), fb1))
        fully1 = tf.nn.dropout(fully1, self.keep_place)

        fw2 = tf.Variable(tf.random_normal(shape=[1024, 1024], stddev=0.01), dtype=tf.float32)
        fb2 = tf.Variable(tf.random_normal(shape=[1024]), dtype=tf.float32)
        fully2 = tf.nn.relu(tf.add(tf.matmul(fully1, fw2), fb2))
        fully2 = tf.nn.dropout(fully2, self.keep_place)

        fw4 = tf.Variable(tf.random_normal(shape=[1024, self.dataset_len], stddev=0.01), dtype=tf.float32)
        fb4 = tf.Variable(tf.random_normal(shape=[self.dataset_len]), dtype=tf.float32)
        fully4 = tf.add(tf.matmul(fully2, fw4), fb4, name='out_put')

识别结果图如图所示：

GUI界面代码如下：

import tkinter as tk
from tkinter.filedialog import *
from tkinter import ttk
import cv2
from PIL import Image, ImageTk
import time
import Apply
import numpy as np
import threading
import tkinter.messagebox as mBox
import App_LPR as al


class Surface(ttk.Frame):
    pic_path = ""
    viewhigh = 300
    viewwide = 300
    update_time = 0
    thread = None
    thread_run = False
    camera = None


    def __init__(self, win):
        ttk.Frame.__init__(self, win)
        frame_left = ttk.Frame(self)
        frame_right1 = ttk.Frame(self)
        frame_right2 = ttk.Frame(self)
        win.title("车牌识别")
        win.state("zoomed")
        self.pack(fill=tk.BOTH, expand=tk.YES, padx="5", pady="5")
        frame_left.pack(side=LEFT,expand=1,fill=BOTH)
        frame_right1.pack(side=TOP,expand=1,fill=tk.Y)
        frame_right2.pack(side=RIGHT,expand=0)
        ttk.Label(frame_left, text='原图：').pack(anchor="nw")
        ttk.Label(frame_right1, text='车牌位置：').grid(column=0, row=0, sticky=tk.W)
        from_pic_ctl = ttk.Button(frame_right2, text="打开车牌图片", width=20, command=self.from_pic)
        self.image_ctl = ttk.Label(frame_left)
        self.image_ctl.pack(anchor="nw")
        self.roi_ctl = ttk.Label(frame_right1)
        self.roi_ctl.grid(column=0, row=1, sticky=tk.W)
        ttk.Label(frame_right1, text='识别结果：').grid(column=0, row=2, sticky=tk.W)
        self.r_ctl = ttk.Label(frame_right1, text="")
        self.r_ctl.grid(column=0, row=3, sticky=tk.W)
        from_pic_ctl.pack(anchor="se", pady="5")

    def get_imgtk(self, img_bgr):
        img = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
        im = Image.fromarray(img)
        imgtk = ImageTk.PhotoImage(image=im) # 文本框里插入图片
        wide = imgtk.width()
        high = imgtk.height()
        if wide > self.viewwide or high > self.viewhigh:
            wide_factor = self.viewwide / wide
            high_factor = self.viewhigh / high
            factor = min(wide_factor, high_factor)
            wide = int(wide * factor)
            if wide <= 0:
                wide = 1
            high = int(high * factor)
            if high <= 0:
                high = 1
            imgtk = ImageTk.PhotoImage(image=im)
        return imgtk

    def show_roi(self, text, roi):
        if text :
            roi = cv2.cvtColor(roi, cv2.COLOR_BGR2RGB)
            roi = Image.fromarray(roi)
            self.imgtk_roi = ImageTk.PhotoImage(image=roi)
            self.roi_ctl.configure(image=self.imgtk_roi, state='enable')
            self.r_ctl.configure(text=str(text))
            self.update_time = time.time()
        elif self.update_time + 8 < time.time():
            self.roi_ctl.configure(state='disabled')
            self.r_ctl.configure(text="")

    def from_pic(self):
        self.thread_run = False
        self.pic_path = askopenfilename(title="选择识别图片", filetypes=[("jpg图片", "*.jpg")])
        img = cv2.imread(self.pic_path )
        img_1 = cv2.imdecode(np.fromfile(self.pic_path, dtype=np.uint8), cv2.IMREAD_COLOR)
        print(self.pic_path)
        if self.pic_path:
            img_bgr = Apply.pre_process(img)
            self.imgtk = self.get_imgtk(img_1) # 将读取的图片插入调整大小并显示
            self.image_ctl.configure(image=self.imgtk)

            # 车牌定位
            car_plate_list = Apply.locate_carPlate(img, img_bgr)

            # 字符提取
            char_img_list = Apply.extract_char(car_plate_list)

            if len(char_img_list) != 7:
                print('传统方法检测失误！！！')
                res_set, image_rgb = al.SimpleRecognizePlateByE2E(img)
                for pstr, confidence, rect in res_set:
                    self.show_roi(pstr, image_rgb)
            else:
                # CNN字符识别
                char_model_path = os.path.join( './model/model.ckpt-1000.meta')
                text = Apply.cnn_recongnize_char(char_img_list, char_model_path)
                self.show_roi(text, car_plate_list)

def close_window():
    print("destroy")
    if surface.thread_run :
        surface.thread_run = False
        surface.thread.join(2.0)
    win.destroy()

if __name__ == '__main__':
    win=tk.Tk()
    surface = Surface(win)
    win.protocol('WM_DELETE_WINDOW', close_window)
    win.mainloop()

整个系统实现了从图像采集，到训练，再到识别的整体流程，想要数据集和代码，v：mql13148