Python_科学计算库
说明：若没有训练级联表，则需要相关级联表才能实现功能
文字识别
# -*- coding: utf-8 -*-
"""
简介：用样本训练数据，再识别
""" import cv2
import numpy as np
from PIL import Image #Python Image Lib
import skimage.feature as feature2d
import sklearn.neighbors as nhb
from sklearn.externals import joblib #对训练模型保存或读取
#cvhog=cv2.HOGDescriptor() #预处理图片
def imgPrepare(filename):
img=cv2.imread(filename,0)
img=np.uint8(img/img.ptp()*255)
img=np.where(img>128,255,img)
img=np.where(img<=128,0,img)
img=np.bitwise_not(img)
return img #横切
def splitchar(img,axis=1):
idxrowb=np.all(img<64,axis=axis)
idxrowi=np.uint8(idxrowb).ravel()
dy=idxrowi[1:]-idxrowi[:-1]
#print(dy)
rowb=np.argwhere(dy==255).ravel()
rowe=np.argwhere(dy==1).ravel()
#print(rowb,rowe)
if axis==1:
imglines=[img[b:e+1,:] for b,e in zip(rowb,rowe)]
else:
imglines=[img[:,b:e+1] for b,e in zip(rowb,rowe)] return imglines #切块
def splitBox(img):
idxrowb=np.all(img<64,axis=1)
idxrowi=np.uint8(idxrowb).ravel()
dy=idxrowi[1:]-idxrowi[:-1]
#print(dy)
rowb=np.argwhere(dy==255).ravel()
rowe=np.argwhere(dy==1).ravel()
b=0
e=-1
if len(rowe)>0:
e=rowe[-1]+1
if len(rowb)>0:
b=rowb[0]
return img[b:e,:] #把图片整成一样大小
def myResize(img,size=(48,48)):
h,w=img.shape
bw=max(h,w)
bh=bw
bimg=np.zeros((bh,bw),np.uint8)
if bw==w:
dh=(bh-h)//2
bimg[dh:dh+h,:]=img[:,:]
else:
dw=(bw-w)//2
bimg[:,dw:dw+w]=img[:,:] bimg=cv2.resize(bimg,size)
return bimg #获取hog向量 图片转为向量
def getHog(img,cell=(16,16),block=(3,3)):
vec=feature2d.hog(img,12,cell,block,'L2')
return vec #训练的主方法
gimg=imgPrepare('e:/sx.jpg')
lines=splitchar(gimg,axis=1)
chars=[]
for line in lines:
charlist=splitchar(line,axis=0)
cchars=[ myResize(splitBox(c)) for c in charlist]
chars.append(cchars)
chars=np.asarray(chars)
X=[]
Y=[]
y=0
for linech in chars: for ch in linech:
chhog=getHog(ch)
X.append(chhog)
Y.append(y) y+=1 KNC=nhb.KNeighborsClassifier(algorithm='ball_tree',n_neighbors=3)
KNC.fit(X,Y) joblib.dump(KNC,'knc.knn') # 识别的主方法
def predict(img):
knc=nhb.KNeighborsClassifier(algorithm='ball_tree',n_neighbors=3)
knc=joblib.load('knc.knn')
lines=splitchar(img,axis=1)
chars=[]
for line in lines:
charlist=splitchar(line,axis=0)
cchars=[ myResize(splitBox(c)) for c in charlist]
chars.append(cchars) chars=np.asarray(chars) Y=[]
for linech in chars:
x=[]
for ch in linech:
chhog=getHog(ch)
x.append(chhog) y=knc.predict(x)
print(y)
Y.append(y) return Y
文字识别
语音处理
def input(self,overtime=60,Noise=12000):
time.sleep(0.5)
pa=au.PyAudio()
stream=pa.open(format = au.paInt16, channels = 1, rate = 16000, input = True,frames_per_buffer = 4000)
spk=pa.open(format=au.paInt16,channels=1,rate=16000,output=True,frames_per_buffer=1000)
filename='./temp/in_%s.wav'%(self._gettoken()) #pcm格式
wf = wave.open(filename, 'wb')
wf.setnchannels(1)
wf.setsampwidth(2)
wf.setframerate(16000)
ch=0
ptparr=np.array([0,0,0,0])
begin=False
while ch<overtime*4:
ch+=1
bs=stream.read(4000)
#spk.write(bs)
arr=np.frombuffer(bs,dtype=np.short)
ptp=arr.max()*1.0-arr.min()*1.0
ptparr[:-1]=ptparr[1:]
ptparr[-1]=np.abs(ptp)
if not begin:
if ptparr[-1]>Noise * 1.5:
begin=True
ch=1
wf.writeframes(bs)
if self.debuge:
print('+',end='')
else:
if np.all(ptparr<Noise):
if self.debuge:
print('+')
break
else:
if self.debuge:
print('-',end='')
wf.writeframes(bs)
stream.close()
spk.close()
wf.close()
wr=wave.open(filename,'rb')
buf=wr.readframes(wr.getnframes())
wr.close()
pa.terminate()
return filename,buf
# self.speech.asr()
def inputvoice(self,overtime=60,Noise=12000):
fn,buf=self.input(overtime,Noise)
result=self.speech.asr(buf)
msgs=[]
if 'result' in result.keys():
msgs=result['result']
msg=''
for m in msgs:
msg+=str(m)
return result['err_no'],msg
语言处理
#语音处理，录音
def input(self,overtime=60,Noise=12000):
time.sleep(0.5)
pa=au.PyAudio()
stream=pa.open(format = au.paInt16, channels = 1, rate = 16000, input = True,frames_per_buffer = 4000)
spk=pa.open(format=au.paInt16,channels=1,rate=16000,output=True,frames_per_buffer=1000)
filename='./temp/in_%s.wav'%(self._gettoken()) #pcm格式
wf = wave.open(filename, 'wb')
wf.setnchannels(1)
wf.setsampwidth(2)
wf.setframerate(16000)
ch=0
ptparr=np.array([0,0,0,0])
begin=False
while ch<overtime*4:
ch+=1
bs=stream.read(4000)
#spk.write(bs)
arr=np.frombuffer(bs,dtype=np.short)
ptp=arr.max()*1.0-arr.min()*1.0
ptparr[:-1]=ptparr[1:]
ptparr[-1]=np.abs(ptp)
if not begin:
if ptparr[-1]>Noise * 1.5:
begin=True
ch=1
wf.writeframes(bs)
if self.debuge:
print('+',end='')
else:
if np.all(ptparr<Noise):
if self.debuge:
print('+')
break
else:
if self.debuge:
print('-',end='')
wf.writeframes(bs)
stream.close()
spk.close()
wf.close()
wr=wave.open(filename,'rb')
buf=wr.readframes(wr.getnframes())
wr.close()
pa.terminate()
return filename,buf
# self.speech.asr()
def inputvoice(self,overtime=60,Noise=12000):
fn,buf=self.input(overtime,Noise)
result=self.speech.asr(buf)
msgs=[]
if 'result' in result.keys():
msgs=result['result']
msg=''
for m in msgs:
msg+=str(m)
return result['err_no'],msg
语音处理（录音）
import cv2
import numpy as np
from PIL import Image
#pip install PIL
#pip install opencv-python
#pip install dlib
dector=cv2.CascadeClassifier()
ret=dector.load('haarcascade_frontalface_alt_tree.xml')
if not ret:
print('未找到级联表文件:plate_cascade.xml')
exit() img=cv2.imread('e:/85n.jpg')
if img is None:
print('文件不存在')
exit()
#彩色转成灰度图像
gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY) gray=np.uint8(gray/gray.ptp()*255) boxs=dector.detectMultiScale(gray,1.015,1)
platelist=[]
for box in boxs:
x,y,w,h=box
g=img[y:y+h,x:x+w,:]
platelist.append(g)
linew=h//100+1
cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),linew)
gimg=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
image=Image.fromarray(gimg)
image.show()
image.close()
人脸识别
import cv2
detector=cv2.CascadeClassifier()
ret=detector.load('plate_cascade.xml')
if not ret:
print('error')
quit()
img=cv2.imread('cars1.jpg')
gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
boxs=detector.detectMultiScale(gray,1.01,3)
for box in boxs:
x,y,w,h=box
p=img[y:y+h,x:x+w:]
name='%d_%d.jpg'%(x,h)
cv2.imwrite(name,p)
车牌识别
# -*- coding: utf-8 -*-
"""
Created on Thu May 17 18:13:35 2018 @author: inspiron
""" import cv2
from PIL import Image hog=cv2.HOGDescriptor()
hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())
img=cv2.imread('e:/1.jpg')
boxs,rets=hog.detectMultiScale(img) for box in boxs:
x,y,w,h=box
cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
gimg=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
image=Image.fromarray(gimg)
image.show() cam=cv2.VideoCapture(0)
while True:
ret,img=cam.read()
if not ret:
break
boxs,rets=hog.detectMultiScale(img)
for box in boxs:
x,y,w,h=box
cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
cv2.imshow('hog',img)
ch=cv2.waitKey(5)
if ch==27:
break cv2.destroyAllWindows()
cam.release()
人形识别
# -*- coding: utf-8 -*-
"""
Created on Thu May 17 19:30:13 2018 @author: AI04班级
""" import cv2
import numpy as np
from PIL import Image #Python Image Lib
import skimage.feature as feature2d
import sklearn.neighbors as nhb
from sklearn.externals import joblib #对训练模型保存或读取
#cvhog=cv2.HOGDescriptor() def imgPrepare(filename):
img=cv2.imread(filename,0)
img=np.uint8(img/img.ptp()*255)
img=np.where(img>128,255,img)
img=np.where(img<=128,0,img)
img=np.bitwise_not(img)
return img def splitchar(img,axis=1):
idxrowb=np.all(img<64,axis=axis)
idxrowi=np.uint8(idxrowb).ravel()
dy=idxrowi[1:]-idxrowi[:-1]
#print(dy)
rowb=np.argwhere(dy==255).ravel()
rowe=np.argwhere(dy==1).ravel()
#print(rowb,rowe)
if axis==1:
imglines=[img[b:e+1,:] for b,e in zip(rowb,rowe)]
else:
imglines=[img[:,b:e+1] for b,e in zip(rowb,rowe)] return imglines def splitBox(img):
idxrowb=np.all(img<64,axis=1)
idxrowi=np.uint8(idxrowb).ravel()
dy=idxrowi[1:]-idxrowi[:-1]
#print(dy)
rowb=np.argwhere(dy==255).ravel()
rowe=np.argwhere(dy==1).ravel()
b=0
e=-1
if len(rowe)>0:
e=rowe[-1]+1
if len(rowb)>0:
b=rowb[0] return img[b:e,:] def myResize(img,size=(48,48)):
h,w=img.shape
bw=max(h,w)
bh=bw
bimg=np.zeros((bh,bw),np.uint8)
if bw==w:
dh=(bh-h)//2
bimg[dh:dh+h,:]=img[:,:]
else:
dw=(bw-w)//2
bimg[:,dw:dw+w]=img[:,:] bimg=cv2.resize(bimg,size)
return bimg def getHog(img,cell=(16,16),block=(3,3)):
vec=feature2d.hog(img,12,cell,block,'L2')
return vec
#main
gimg=imgPrepare('e:/sx.jpg')
lines=splitchar(gimg,axis=1)
chars=[]
for line in lines:
charlist=splitchar(line,axis=0)
cchars=[ myResize(splitBox(c)) for c in charlist]
chars.append(cchars)
chars=np.asarray(chars)
X=[]
Y=[]
y=0
for linech in chars: for ch in linech:
chhog=getHog(ch)
X.append(chhog)
Y.append(y) y+=1 KNC=nhb.KNeighborsClassifier(algorithm='ball_tree',n_neighbors=3)
KNC.fit(X,Y) joblib.dump(KNC,'knc.knn') def predict(img):
knc=nhb.KNeighborsClassifier(algorithm='ball_tree',n_neighbors=3)
knc=joblib.load('knc.knn')
lines=splitchar(img,axis=1)
chars=[]
for line in lines:
charlist=splitchar(line,axis=0)
cchars=[ myResize(splitBox(c)) for c in charlist]
chars.append(cchars) chars=np.asarray(chars) Y=[]
for linech in chars:
x=[]
for ch in linech:
chhog=getHog(ch)
x.append(chhog) y=knc.predict(x)
print(y)
Y.append(y) return Y
数字识别
Python_科学计算库的更多相关文章
- Python_科学计算平台__pypi体系的numpy、scipy、pandas、matplotlib库简介
1.numpy--基础,以矩阵为基础的数学计算模块,纯数学 存储和处理大型矩阵. 这个是很基础的扩展,其余的扩展都是以此为基础. 快速学习入口 https://docs.scipy.org/doc/n ...
- SciPy - 科学计算库（上）
SciPy - 科学计算库(上) 一.实验说明 SciPy 库建立在 Numpy 库之上,提供了大量科学算法,主要包括这些主题: 特殊函数 (scipy.special) 积分 (scipy.inte ...
- python科学计算库的numpy基础知识，完美抽象多维数组(原创)
#导入科学计算库 #起别名避免重名 import numpy as np #小技巧:从外往内看==从左往右看 从内往外看==从右往左看 #打印版本号 print(np.version.version) ...
- python科学计算库numpy和绘图库PIL的结合,素描图片(原创)
# 导入绘图库 from PIL import Image #导入科学计算库 import numpy as np #封装一个图像处理工具类 class TestNumpy(object): def ...
- numpy科学计算库的基础用法，完美抽象多维数组(原创)
#起别名避免重名 import numpy as np #小技巧:print从外往内看==shape从左往右看 if __name__ == "__main__": print(' ...
- Python科学计算库
Python科学计算库 一.numpy库和matplotlib库的学习 (1)numpy库介绍:科学计算包,支持N维数组运算.处理大型矩阵.成熟的广播函数库.矢量运算.线性代数.傅里叶变换.随机数生成 ...
- ubuntu14.04 下安装 gsl 科学计算库
GSL(GNU Scientific Library)作为三大科学计算库之一,除了涵盖基本的线性代数,微分方程,积分,随机数,组合数,方程求根,多项式求根,排序等,还有模拟退火,快速傅里叶变换,小波, ...
- windows下如何快速优雅的使用python的科学计算库？
Python是一种强大的编程语言,其提供了很多用于科学计算的模块,常见的包括numpy.scipy.pandas和matplotlib.要利用Python进行科学计算,就需要一一安装所需的模块,而这些 ...
- Python科学计算库Numpy
Python科学计算库Numpy NumPy(Numerical Python) 是 Python 语言的一个扩展程序库,支持大量的维度数组与矩阵运算,此外也针对数组运算提供大量的数学函数库. 1.简 ...
随机推荐
- mongodb安装及使用
安装命令: sudo apt-get install mongodb 开始认证,创建用户: 编辑配置文件: sudo vim /etc/mongodb.conf 11行中的 bind_ip值 修改成为 ...
- Harbor 安装教程
Harbor 安装教程 一. CentOS设置 1. 更换阿里源 curl -o /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com ...
- 子网划分和VLAN
子网划分 IP地址的结构和分类 根据tcp/ip协议,连接在Internet上的每个设备都必须有一个IP地址,它是一个32位二进制数,为了方便人类识别,我们将它用点分十进制表示,每8位分为一段. IP ...
- poj1050 To the Max（降维dp）
To the Max Time Limit: 1000MS Memory Limit: 10000K Total Submissions: 49351 Accepted: 26142 Desc ...
- Dockerfile 笔记
Dockerfile ARGARG <name>[=<default value>]The ARG instruction defines a variable that ...
- mysql 架构简介
mysql的逻辑架构 第一层:进行连接处理.权限认证.安全校验等. 当客户端(应用)连接到mysql服务器时,服务器会创建使用一个线程进行处理连接(少量的线程服务大量的连接),随后服务器需要对该连接进 ...
- 老板，来几道web玩玩
好久没做web了,没想到还能自己做出来555 [MRCTF2020]Ez_bypass 签到题8 给了源码,一个md5强类型比较,然后post传参,弱类型判断,直接1234567a绕过了 I put ...
- Altium Designer中如何批量修改元器件封装？
我想你说的应该是altium里的封装管理库吧.1,Tools -> Footprint Manager -> ...2,在Component List里选择要改的器件3,在View and ...
- NB-IoT和LORA技术通信距离是一样的吗
如今物联网的无线通信技术非常多,这其中主要分为两大类:一类是以Zigbee.WiFi.蓝牙.Z-wave等短距离通信技术为主:另一类就是以LPWAN(low-powerWide-AreaNetwork ...
- LoRa技术的发展应用和LoRa应用设备
LoRa技术的发展应用 LORA技术大约在十年前由法国和瑞士开发,到现今LORA技术已经是物联网发展应用中不可缺少的一部分,根据中国物联网研究与发展中心的数据,2025年我国物联网产业规模将达到2万亿 ...