imageClassification 基于朴素贝叶斯

最新推荐文章于 2022-05-30 09:21:53 发布

偷嘴的小猴子

最新推荐文章于 2022-05-30 09:21:53 发布

阅读量475

点赞数

分类专栏： ImageClassification 文章标签： imageclassification python bayer

本文链接：https://blog.csdn.net/a793620101/article/details/51175337

版权

ImageClassification 专栏收录该内容

5 篇文章 0 订阅

订阅专栏

本文介绍了如何使用朴素贝叶斯方法进行图像分类，特别是针对手写数字图像。通过将每个像素作为特征，并计算先验概率和似然概率来实现分类。详细内容包括原理解释、实验介绍以及数据来源。

摘要由CSDN通过智能技术生成

一.原理

朴素贝叶斯是一种构建分类器的简单方法。对于某些类型的概率模型，在监督式学习的样本集中能获取得非常好的分类效果。朴素贝叶斯方法常用于文本分类，这次我们将使用贝叶斯方法进行图像分类。贝叶斯方法的原理，可以使用一个公式表达：

详细内容可参考：贝叶斯分类器

二.介绍

这次实验主要是实现手写数字图像的分类。我们把每一个像素都作为图像的特征，Fij表示像素(x,y)的值，把每个类的数目和总的图像的比值作为先验概率，而似然概率：

后验概率

P(class) ⋅ P(f_1,1 | class) ⋅ P(f_1,2 | class) ⋅ ... ⋅ P(f_28,28 | class).

log P(class) + log P(f_1,1 | class) + log P(f_1,2 | class) + ... + log P(f_28,28 | class).

图像形式：

00000000000001111000000000000000
00000000000011111110000000000000
00000000001111111111000000000000
00000001111111111111100000000000
00000001111111011111100000000000
00000011111110000011110000000000
00000011111110000000111000000000
00000011111110000000111100000000
00000011111110000000011100000000
00000011111110000000011100000000
00000011111100000000011110000000
00000011111100000000001110000000
00000011111100000000001110000000
00000001111110000000000111000000
00000001111110000000000111000000
00000001111110000000000111000000
00000001111110000000000111000000
00000011111110000000001111000000
00000011110110000000001111000000
00000011110000000000011110000000
00000001111000000000001111000000
00000001111000000000011111000000
00000001111000000000111110000000
00000001111000000001111100000000
00000000111000000111111000000000
00000000111100011111110000000000
00000000111111111111110000000000
00000000011111111111110000000000
00000000011111111111100000000000
00000000001111111110000000000000
00000000000111110000000000000000
00000000000011000000000000000000

数据下载：这里

三.实现

from numpy import *
import os

def img2matrix(filename):
	im=open(filename)
	m=32
	mat=zeros((1,1024))
	for i in range(m):
		lineStr=im.readline()
		for j in range(m):
			mat[0,32*i+j]=int(lineStr[j])
	return mat

def file2data():
	trainlabel=[]
	testlabel=[]
	dir='D:\\digits'
	dir1=os.path.join(dir,'trainingDigits')
	traindata=zeros((len(os.listdir(dir1)),1024))
	dir2=os.path.join(dir,'testDigits')
	testdata=zeros((len(os.listdir(dir2)),1024))
	x=0
	for i in os.listdir(dir1):
		filename=i.split('.')[0]
		l=int(filename.split('_')[0])
		trainlabel.append(l)
		f=os.path.join(dir1,i)
		traindata[x,:]=img2matrix(f)
		x=x+1
		
	x=0
	for i in os.listdir(dir2):
		filename=i.split('.')[0]
		l=int(filename.split('_')[0])
		testlabel.append(l)
		f=os.path.join(dir2,i)
		testdata[x,:]=img2matrix(f)
		x+=1
	return traindata,trainlabel,testdata,testlabel

def training(traindata,trainlabel):
	dict1={}
	prior=zeros((10,1))
	trainall=len(trainlabel)
	for i in trainlabel:
		if i not in dict1:
			dict1[i]=1
		else:
			dict1[i]+=1
	for key,value in dict1.items():
		prior[key,0]=1.0*value/trainall

	p=ones((10,1024))
	z=0
	s=0
	for key,value in dict1.items():
		m=value
		for j in range(1024):
			for i in range(s,s+value):
				if traindata[i,j]==0:
					z+=1
			p[key,j]=1.0*z/(1+m)
			z=0

		s+=value
	return dict1,prior,p

def testing(testdata,prior,p):
	predict=zeros((10,1))
	tmp=0.0
	t=1.0
	for i in range(10):
		for j in range(1024):
			if testdata[j]==0.0:
				tmp=p[i,j]
			else:
				tmp=1-p[i,j]
			t=t*tmp
		predict[i,0]=t*prior[i,0]
		t=1.0
	pt=argsort(-predict,axis=0)  #descending sort
	return pt[0]

def bayerClassifier(testdata,testlabel,prior,p):
	m=len(testdata)
	count=0
	for i in range(m):
		c=testing(testdata[i,:],prior,p)
		if c==testlabel[i]:
			count+=1

		print 'the predict class is: %d, the real class is: %d' %(c,testlabel[i])
	print 'The accuracy is: %0.2f%%'% (100.0*count/m)


if __name__=='__main__':
	print '********Preparing data*****'
	traindata,trainlabel,testdata,testlabel=file2data()
	print '********Training***********'
	dict1,prior,p=training(traindata,trainlabel)
	print '********Testing************'
	bayerClassifier(testdata,testlabel,prior,p)

四.结果

********Preparing data*****
********Training***********
********Testing************
the predict class is: 0, the real class is: 0
the predict class is: 0, the real class is: 0
...
...
...
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
the predict class is: 9, the real class is: 9
The accuracy is: 92.18%
[Finished in 7.9s]