【深度学习系列】用PaddlePaddle进行车牌识别（一）
小伙伴们，终于到了实战部分了！今天给大家带来的项目是用PaddlePaddle进行车牌识别。车牌识别其实属于比较常见的图像识别的项目了，目前也属于比较成熟的应用，大多数老牌厂家能做到准确率99%+。传统的方法需要对图像进行多次预处理再用机器学习的分类算法进行分类识别，然而深度学习发展起来以后，我们可以通过用CNN来进行端对端的车牌识别。任何模型的训练都离不开数据，在车牌识别中，除了晚上能下载到的一些包含车牌的数据是不够的，本篇文章的主要目的是教大家如何批量生成车牌。
生成车牌数据
1.定义车牌数据所需字符
车牌中包括省份简称、大写英文字母和数字，我们首先定义需要的字符和字典，方便后面使用
index = {"京": 0, "沪": 1, "津": 2, "渝": 3, "冀": 4, "晋": 5, "蒙": 6, "辽": 7, "吉": 8, "黑": 9, "苏": 10, "浙": 11, "皖": 12,
"闽": 13, "赣": 14, "鲁": 15, "豫": 16, "鄂": 17, "湘": 18, "粤": 19, "桂": 20, "琼": 21, "川": 22, "贵": 23, "云": 24,
"藏": 25, "陕": 26, "甘": 27, "青": 28, "宁": 29, "新": 30, "": 31, "": 32, "": 33, "": 34, "": 35, "": 36,
"": 37, "": 38, "": 39, "": 40, "A": 41, "B": 42, "C": 43, "D": 44, "E": 45, "F": 46, "G": 47, "H": 48,
"J": 49, "K": 50, "L": 51, "M": 52, "N": 53, "P": 54, "Q": 55, "R": 56, "S": 57, "T": 58, "U": 59, "V": 60,
"W": 61, "X": 62, "Y": 63, "Z": 64};
chars = ["京", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙", "皖", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤", "桂",
"琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁", "新", "", "", "", "", "", "", "", "", "", "", "A",
"B", "C", "D", "E", "F", "G", "H", "J", "K", "L", "M", "N", "P", "Q", "R", "S", "T", "U", "V", "W", "X",
"Y", "Z"
];
2.生成中英文字符
def GenCh(f,val):
"""
生成中文字符
"""
img=Image.new("RGB", (45,70),(255,255,255))
draw = ImageDraw.Draw(img)
draw.text((0, 3),val,(0,0,0),font=f)
img = img.resize((23,70))
A = np.array(img)
return A def GenCh1(f,val):
"""
生成英文字符
"""
img=Image.new("RGB", (23,70),(255,255,255))
draw = ImageDraw.Draw(img)
draw.text((0, 2),val.decode('utf-8'),(0,0,0),font=f)
A = np.array(img)
return A
3.对数据添加各种噪音和畸变，模糊处理
def AddSmudginess(img, Smu):
rows = r(Smu.shape[0] - 50)
cols = r(Smu.shape[1] - 50)
adder = Smu[rows:rows + 50, cols:cols + 50];
adder = cv2.resize(adder, (50, 50));
#adder = cv2.bitwise_not(adder)
img = cv2.resize(img,(50,50))
img = cv2.bitwise_not(img)
img = cv2.bitwise_and(adder, img)
img = cv2.bitwise_not(img)
return img def rot(img,angel,shape,max_angel):
"""
添加放射畸变
img 输入图像
factor 畸变的参数
size 为图片的目标尺寸
"""
size_o = [shape[1],shape[0]]
size = (shape[1]+ int(shape[0]*cos((float(max_angel )/180) * 3.14)),shape[0])
interval = abs( int( sin((float(angel) /180) * 3.14)* shape[0]));
pts1 = np.float32([[0,0],[0,size_o[1]],[size_o[0],0],[size_o[0],size_o[1]]])
if(angel>0):
pts2 = np.float32([[interval,0],[0,size[1] ],[size[0],0 ],[size[0]-interval,size_o[1]]])
else:
pts2 = np.float32([[0,0],[interval,size[1] ],[size[0]-interval,0 ],[size[0],size_o[1]]])
M = cv2.getPerspectiveTransform(pts1,pts2);
dst = cv2.warpPerspective(img,M,size);
return dst def rotRandrom(img, factor, size):
"""
添加透视畸变
"""
shape = size;
pts1 = np.float32([[0, 0], [0, shape[0]], [shape[1], 0], [shape[1], shape[0]]])
pts2 = np.float32([[r(factor), r(factor)], [ r(factor), shape[0] - r(factor)], [shape[1] - r(factor), r(factor)],
[shape[1] - r(factor), shape[0] - r(factor)]])
M = cv2.getPerspectiveTransform(pts1, pts2);
dst = cv2.warpPerspective(img, M, size);
return dst def tfactor(img):
"""
添加饱和度光照的噪声
"""
hsv = cv2.cvtColor(img,cv2.COLOR_BGR2HSV);
hsv[:,:,0] = hsv[:,:,0]*(0.8+ np.random.random()*0.2);
hsv[:,:,1] = hsv[:,:,1]*(0.3+ np.random.random()*0.7);
hsv[:,:,2] = hsv[:,:,2]*(0.2+ np.random.random()*0.8); img = cv2.cvtColor(hsv,cv2.COLOR_HSV2BGR);
return img def random_envirment(img,data_set):
"""
添加自然环境的噪声
"""
index=r(len(data_set))
env = cv2.imread(data_set[index])
env = cv2.resize(env,(img.shape[1],img.shape[0]))
bak = (img==0);
bak = bak.astype(np.uint8)*255;
inv = cv2.bitwise_and(bak,env)
img = cv2.bitwise_or(inv,img)
return img def AddGauss(img, level):
"""
添加高斯模糊
"""
return cv2.blur(img, (level * 2 + 1, level * 2 + 1)); def r(val):
return int(np.random.random() * val) def AddNoiseSingleChannel(single):
"""
添加高斯噪声
"""
diff = 255-single.max();
noise = np.random.normal(0,1+r(6),single.shape);
noise = (noise - noise.min())/(noise.max()-noise.min())
noise= diff*noise;
noise= noise.astype(np.uint8)
dst = single + noise
return dst def addNoise(img,sdev = 0.5,avg=10):
img[:,:,0] = AddNoiseSingleChannel(img[:,:,0]);
img[:,:,1] = AddNoiseSingleChannel(img[:,:,1]);
img[:,:,2] = AddNoiseSingleChannel(img[:,:,2]);
return img
4.加入背景图片，生成车牌字符串list和label，并存为图片格式，批量生成。
class GenPlate:
def __init__(self,fontCh,fontEng,NoPlates):
self.fontC = ImageFont.truetype(fontCh,43,0);
self.fontE = ImageFont.truetype(fontEng,60,0);
self.img=np.array(Image.new("RGB", (226,70),(255,255,255)))
self.bg = cv2.resize(cv2.imread("./images/template.bmp"),(226,70));
self.smu = cv2.imread("./images/smu2.jpg");
self.noplates_path = [];
for parent,parent_folder,filenames in os.walk(NoPlates):
for filename in filenames:
path = parent+"/"+filename;
self.noplates_path.append(path);
def draw(self,val):
offset= 2 ;
self.img[0:70,offset+8:offset+8+23]= GenCh(self.fontC,val[0]);
self.img[0:70,offset+8+23+6:offset+8+23+6+23]= GenCh1(self.fontE,val[1]);
for i in range(5):
base = offset+8+23+6+23+17 +i*23 + i*6 ;
self.img[0:70, base : base+23]= GenCh1(self.fontE,val[i+2]);
return self.img
def generate(self,text):
if len(text) == 9:
fg = self.draw(text.decode(encoding="utf-8"));
fg = cv2.bitwise_not(fg);
com = cv2.bitwise_or(fg,self.bg);
com = rot(com,r(60)-30,com.shape,30);
com = rotRandrom(com,10,(com.shape[1],com.shape[0]));
com = tfactor(com)
com = random_envirment(com,self.noplates_path);
com = AddGauss(com, 1+r(4));
com = addNoise(com);
return com
def genPlateString(self,pos,val):
'''
生成车牌String,存为图片
生成车牌list,存为label
'''
plateStr = "";
plateList=[]
box = [0,0,0,0,0,0,0];
if(pos!=-1):
box[pos]=1;
for unit,cpos in zip(box,range(len(box))):
if unit == 1:
plateStr += val
#print plateStr
plateList.append(val)
else:
if cpos == 0:
plateStr += chars[r(31)]
plateList.append(plateStr)
elif cpos == 1:
plateStr += chars[41+r(24)]
plateList.append(plateStr)
else:
plateStr += chars[31 + r(34)]
plateList.append(plateStr)
plate = [plateList[0]]
b = [plateList[i][-1] for i in range(len(plateList))]
plate.extend(b[1:7])
return plateStr,plate
# 将生成的车牌图片写入文件夹，对应的label写入label.txt
def genBatch(self, batchSize,pos,charRange, outputPath,size):
if (not os.path.exists(outputPath)):
os.mkdir(outputPath)
outfile = open('label.txt','w')
for i in xrange(batchSize):
plateStr,plate = G.genPlateString(-1,-1)
print plateStr,plate
img = G.generate(plateStr);
img = cv2.resize(img,size);
cv2.imwrite(outputPath + "/" + str(i).zfill(2) + ".jpg", img);
outfile.write(str(plate)+"\n")
G = GenPlate("./font/platech.ttf",'./font/platechar.ttf',"./NoPlates")
完整代码：
#coding=utf-8
"""
genPlate.py:生成随机车牌
""" __author__ = "Huxiaoman"
__copyright__ = "Copyright (c) 2017 " import PIL
from PIL import ImageFont
from PIL import Image
from PIL import ImageDraw
import cv2;
import numpy as np;
import os;
from math import *
import sys index = {"京": 0, "沪": 1, "津": 2, "渝": 3, "冀": 4, "晋": 5, "蒙": 6, "辽": 7, "吉": 8, "黑": 9, "苏": 10, "浙": 11, "皖": 12,
"闽": 13, "赣": 14, "鲁": 15, "豫": 16, "鄂": 17, "湘": 18, "粤": 19, "桂": 20, "琼": 21, "川": 22, "贵": 23, "云": 24,
"藏": 25, "陕": 26, "甘": 27, "青": 28, "宁": 29, "新": 30, "": 31, "": 32, "": 33, "": 34, "": 35, "": 36,
"": 37, "": 38, "": 39, "": 40, "A": 41, "B": 42, "C": 43, "D": 44, "E": 45, "F": 46, "G": 47, "H": 48,
"J": 49, "K": 50, "L": 51, "M": 52, "N": 53, "P": 54, "Q": 55, "R": 56, "S": 57, "T": 58, "U": 59, "V": 60,
"W": 61, "X": 62, "Y": 63, "Z": 64}; chars = ["京", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙", "皖", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤", "桂",
"琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁", "新", "", "", "", "", "", "", "", "", "", "", "A",
"B", "C", "D", "E", "F", "G", "H", "J", "K", "L", "M", "N", "P", "Q", "R", "S", "T", "U", "V", "W", "X",
"Y", "Z"
]; def AddSmudginess(img, Smu):
rows = r(Smu.shape[0] - 50)
cols = r(Smu.shape[1] - 50)
adder = Smu[rows:rows + 50, cols:cols + 50];
adder = cv2.resize(adder, (50, 50));
#adder = cv2.bitwise_not(adder)
img = cv2.resize(img,(50,50))
img = cv2.bitwise_not(img)
img = cv2.bitwise_and(adder, img)
img = cv2.bitwise_not(img)
return img def rot(img,angel,shape,max_angel):
"""
添加放射畸变
img 输入图像
factor 畸变的参数
size 为图片的目标尺寸
"""
size_o = [shape[1],shape[0]]
size = (shape[1]+ int(shape[0]*cos((float(max_angel )/180) * 3.14)),shape[0])
interval = abs( int( sin((float(angel) /180) * 3.14)* shape[0]));
pts1 = np.float32([[0,0],[0,size_o[1]],[size_o[0],0],[size_o[0],size_o[1]]])
if(angel>0):
pts2 = np.float32([[interval,0],[0,size[1] ],[size[0],0 ],[size[0]-interval,size_o[1]]])
else:
pts2 = np.float32([[0,0],[interval,size[1] ],[size[0]-interval,0 ],[size[0],size_o[1]]])
M = cv2.getPerspectiveTransform(pts1,pts2);
dst = cv2.warpPerspective(img,M,size);
return dst def rotRandrom(img, factor, size):
"""
添加透视畸变
"""
shape = size;
pts1 = np.float32([[0, 0], [0, shape[0]], [shape[1], 0], [shape[1], shape[0]]])
pts2 = np.float32([[r(factor), r(factor)], [ r(factor), shape[0] - r(factor)], [shape[1] - r(factor), r(factor)],
[shape[1] - r(factor), shape[0] - r(factor)]])
M = cv2.getPerspectiveTransform(pts1, pts2);
dst = cv2.warpPerspective(img, M, size);
return dst def tfactor(img):
"""
添加饱和度光照的噪声
"""
hsv = cv2.cvtColor(img,cv2.COLOR_BGR2HSV);
hsv[:,:,0] = hsv[:,:,0]*(0.8+ np.random.random()*0.2);
hsv[:,:,1] = hsv[:,:,1]*(0.3+ np.random.random()*0.7);
hsv[:,:,2] = hsv[:,:,2]*(0.2+ np.random.random()*0.8); img = cv2.cvtColor(hsv,cv2.COLOR_HSV2BGR);
return img def random_envirment(img,data_set):
"""
添加自然环境的噪声
"""
index=r(len(data_set))
env = cv2.imread(data_set[index])
env = cv2.resize(env,(img.shape[1],img.shape[0]))
bak = (img==0);
bak = bak.astype(np.uint8)*255;
inv = cv2.bitwise_and(bak,env)
img = cv2.bitwise_or(inv,img)
return img def GenCh(f,val):
"""
生成中文字符
"""
img=Image.new("RGB", (45,70),(255,255,255))
draw = ImageDraw.Draw(img)
draw.text((0, 3),val,(0,0,0),font=f)
img = img.resize((23,70))
A = np.array(img)
return A def GenCh1(f,val):
"""
生成英文字符
"""
img=Image.new("RGB", (23,70),(255,255,255))
draw = ImageDraw.Draw(img)
draw.text((0, 2),val.decode('utf-8'),(0,0,0),font=f)
A = np.array(img)
return A def AddGauss(img, level):
"""
添加高斯模糊
"""
return cv2.blur(img, (level * 2 + 1, level * 2 + 1)); def r(val):
return int(np.random.random() * val) def AddNoiseSingleChannel(single):
"""
添加高斯噪声
"""
diff = 255-single.max();
noise = np.random.normal(0,1+r(6),single.shape);
noise = (noise - noise.min())/(noise.max()-noise.min())
noise= diff*noise;
noise= noise.astype(np.uint8)
dst = single + noise
return dst def addNoise(img,sdev = 0.5,avg=10):
img[:,:,0] = AddNoiseSingleChannel(img[:,:,0]);
img[:,:,1] = AddNoiseSingleChannel(img[:,:,1]);
img[:,:,2] = AddNoiseSingleChannel(img[:,:,2]);
return img class GenPlate: def __init__(self,fontCh,fontEng,NoPlates):
self.fontC = ImageFont.truetype(fontCh,43,0);
self.fontE = ImageFont.truetype(fontEng,60,0);
self.img=np.array(Image.new("RGB", (226,70),(255,255,255)))
self.bg = cv2.resize(cv2.imread("./images/template.bmp"),(226,70));
self.smu = cv2.imread("./images/smu2.jpg");
self.noplates_path = [];
for parent,parent_folder,filenames in os.walk(NoPlates):
for filename in filenames:
path = parent+"/"+filename;
self.noplates_path.append(path); def draw(self,val):
offset= 2 ;
self.img[0:70,offset+8:offset+8+23]= GenCh(self.fontC,val[0]);
self.img[0:70,offset+8+23+6:offset+8+23+6+23]= GenCh1(self.fontE,val[1]);
for i in range(5):
base = offset+8+23+6+23+17 +i*23 + i*6 ;
self.img[0:70, base : base+23]= GenCh1(self.fontE,val[i+2]);
return self.img def generate(self,text):
if len(text) == 9:
fg = self.draw(text.decode(encoding="utf-8"));
fg = cv2.bitwise_not(fg);
com = cv2.bitwise_or(fg,self.bg);
com = rot(com,r(60)-30,com.shape,30);
com = rotRandrom(com,10,(com.shape[1],com.shape[0]));
com = tfactor(com)
com = random_envirment(com,self.noplates_path);
com = AddGauss(com, 1+r(4));
com = addNoise(com);
return com def genPlateString(self,pos,val):
'''
生成车牌String,存为图片
生成车牌list,存为label
'''
plateStr = "";
plateList=[]
box = [0,0,0,0,0,0,0];
if(pos!=-1):
box[pos]=1;
for unit,cpos in zip(box,range(len(box))):
if unit == 1:
plateStr += val
#print plateStr
plateList.append(val)
else:
if cpos == 0:
plateStr += chars[r(31)]
plateList.append(plateStr)
elif cpos == 1:
plateStr += chars[41+r(24)]
plateList.append(plateStr)
else:
plateStr += chars[31 + r(34)]
plateList.append(plateStr)
plate = [plateList[0]]
b = [plateList[i][-1] for i in range(len(plateList))]
plate.extend(b[1:7])
return plateStr,plate # 将生成的车牌图片写入文件夹，对应的label写入label.txt
def genBatch(self, batchSize,pos,charRange, outputPath,size):
if (not os.path.exists(outputPath)):
os.mkdir(outputPath)
outfile = open('label.txt','w')
for i in xrange(batchSize):
plateStr,plate = G.genPlateString(-1,-1)
print plateStr,plate
img = G.generate(plateStr);
img = cv2.resize(img,size);
cv2.imwrite(outputPath + "/" + str(i).zfill(2) + ".jpg", img);
outfile.write(str(plate)+"\n")
G = GenPlate("./font/platech.ttf",'./font/platechar.ttf',"./NoPlates")
#G.genBatch(100,2,range(31,65),"./plate_100",(272,72)) if __name__=='__main__':
G.genBatch(int(sys.argv[1]),2,range(31,65),sys.argv[2],(272,72))
运行时加生成数量和保存路径即可，如：
python genPlate.py 100 ./plate_100
显示结果：
上图即为生成的车牌数据，有清晰的有模糊的，有比较方正的，也有一些比较倾斜，生成完大量的车牌样张后就可以进行车牌识别了。下一小节将会讲如何用端对端的CNN进行车牌识别，不需要通过传统的ocr先对字符进行分割处理后再识别。
参考资料：
1.原来做的车牌识别项目：https://github.com/huxiaoman7/mxnet-cnn-plate-recognition
【深度学习系列】用PaddlePaddle进行车牌识别（一）的更多相关文章
- 【深度学习】用PaddlePaddle进行车牌识别（二）
上节我们讲了第一部分,如何用生成简易的车牌,这节课中我们会用PaddlePaddle来识别生成的车牌. 数据读取 在上一节生成车牌时,我们可以分别生成训练数据和测试数据,方法如下(完整代码在这里): ...
- 【深度学习系列】PaddlePaddle垃圾邮件处理实战（二）
PaddlePaddle垃圾邮件处理实战(二) 前文回顾 在上篇文章中我们讲了如何用支持向量机对垃圾邮件进行分类,auc为73.3%,本篇讲继续讲如何用PaddlePaddle实现邮件分类,将深度 ...
- 【深度学习系列】PaddlePaddle之手写数字识别
上周在搜索关于深度学习分布式运行方式的资料时,无意间搜到了paddlepaddle,发现这个框架的分布式训练方案做的还挺不错的,想跟大家分享一下.不过呢,这块内容太复杂了,所以就简单的介绍一下padd ...
- 【深度学习系列】手写数字识别卷积神经--卷积神经网络CNN原理详解(一)
上篇文章我们给出了用paddlepaddle来做手写数字识别的示例,并对网络结构进行到了调整,提高了识别的精度.有的同学表示不是很理解原理,为什么传统的机器学习算法,简单的神经网络(如多层感知机)都可 ...
- 【深度学习系列】PaddlePaddle垃圾邮件处理实战（一）
PaddlePaddle垃圾邮件处理实战(一) 背景介绍 在我们日常生活中,经常会受到各种垃圾邮件,譬如来自商家的广告.打折促销信息.澳门博彩邮件.理财推广信息等,一般来说邮件客户端都会设置一定的 ...
- 【深度学习系列】PaddlePaddle之数据预处理
上篇文章讲了卷积神经网络的基本知识,本来这篇文章准备继续深入讲CNN的相关知识和手写CNN,但是有很多同学跟我发邮件或私信问我关于PaddlePaddle如何读取数据.做数据预处理相关的内容.网上看的 ...
- 【深度学习系列】PaddlePaddle可视化之VisualDL
上篇文章我们讲了如何对模型进行可视化,用的keras手动绘图输出CNN训练的中途结果,本篇文章将讲述如何用PaddlePaddle新开源的VisualDL来进行可视化.在讲VisualDL之前,我们先 ...
- 【AI开发】基于深度学习的卡口车型、车牌识别
服务端代码后面给出 卡口车型.车牌识别demo截图 服务器:
- 【深度学习系列】关于PaddlePaddle的一些避“坑”技巧
最近除了工作以外,业余在参加Paddle的AI比赛,在用Paddle训练的过程中遇到了一些问题,并找到了解决方法,跟大家分享一下: PaddlePaddle的Anaconda的兼容问题 之前我是在服务 ...
随机推荐
- ios 一步一步学会自定义地图吹出框(CalloutView)-->(百度地图，高德地图,google地图)
前言 在 ios上边使用地图库的同学肯定遇到过这样的问题:吹出框只能设置title和subtitle和左右的view,不管是百度地图还是高德地图还是自带的 google地图,只提供了这四个属性,如果想 ...
- vue中引入jQuery和bootstrap
一.引入jQuery: 首先在当前项目的根目录下(就是与package.json同目录),运行命令npm install jquery --save-dev 这样就将jquery安装到了这个项目中 ...
- BIGIP-LTM中的NAT和SNAT
http://250688049.blog.51cto.com/643101/1095880 一.NAT(Network Address Translation)网络地址转换1.NAT简介 NAT ...
- np.array转换为list，嵌套的python list转成一个一维的python list
np.array转换为list 1 meitan = shuju.iloc[start:end, 1:2] zhengqi = shuju.iloc[start:end,2:3] print(type ...
- x64内核HOOK技术之拦截进程.拦截线程.拦截模块
x64内核HOOK技术之拦截进程.拦截线程.拦截模块 一丶为什么讲解HOOK技术. 在32系统下, 例如我们要HOOK SSDT表,那么直接讲CR0的内存保护属性去掉. 直接讲表的地址修改即可. 但是 ...
- 新建maven项目，JRE System Library[J2SE-1.5]
上篇博文中搭建了maven多模块项目,发现全是JRE System Library[J2SE-1.5],如图. 怎么避免这种情况呢? windows-preferences-maven-user se ...
- Python CRM项目七
仿照Django Admin实现对readonly的字段进行设置 功能点: 1.页面不可进行更改 2.如果改变html代码中的值,则需要进行后端的数据库数据校验 3.可以对某些字段进行自定制校验规则 ...
- ABP官方文档翻译 3.1 实体
实体 实体类 聚合根类 领域事件 常规接口 审计 软删除 激活/失活实体 实体改变事件 IEntity接口 实体是DDD(领域驱动设计)的核心概念之一.Eric Evans描述它为"An o ...
- xBIM IFC 墙壁案例
目录 xBIM 应用与学习 (一) xBIM 应用与学习 (二) xBIM 基本的模型操作 xBIM 日志操作 XBIM 3D 墙壁案例 xBIM 格式之间转换 xBIM 使用Linq 来优化查询 x ...
- BZOJ 3473: 字符串 [广义后缀自动机]
3473: 字符串 Time Limit: 20 Sec Memory Limit: 256 MBSubmit: 354 Solved: 160[Submit][Status][Discuss] ...