Python knn 对手写数字分类

淡淡的烟草味﹌ 2022-06-06 12:15 115阅读 0赞

一不小心入了图像处理的大坑，我要一步一步向上爬~

因为本文是直接调用sklearn库中已封装好的KNN，所以对KNN算法就不作介绍了  
\#\#\#1、MNIST数据集转化为图片  
\#\#\#\#\#这部分代码直接从[http://blog.csdn.net/qq\_32166627/article/details/52640730][http_blog.csdn.net_qq_32166627_article_details_52640730]找的  
我们只需要训练集

import numpy as np  
    import struct  
      
    from PIL import Image  
    import os  
      
    data_file = 'somePath/train-images.idx3-ubyte' #需要修改的路径  
    # It's 47040016B, but we should set to 47040000B  
    data_file_size = 47040016  
    data_file_size = str(data_file_size - 16) + 'B'  
      
    data_buf = open(data_file, 'rb').read()  
      
    magic, numImages, numRows, numColumns = struct.unpack_from(  
        '>IIII', data_buf, 0)  
    datas = struct.unpack_from(  
        '>' + data_file_size, data_buf, struct.calcsize('>IIII'))  
    datas = np.array(datas).astype(np.uint8).reshape(  
        numImages, 1, numRows, numColumns)  
      
    label_file = 'somePath/train-labels.idx1-ubyte' #需要修改的路径  
      
    # It's 60008B, but we should set to 60000B  
    label_file_size = 60008  
    label_file_size = str(label_file_size - 8) + 'B'  
      
    label_buf = open(label_file, 'rb').read()  
      
    magic, numLabels = struct.unpack_from('>II', label_buf, 0)  
    labels = struct.unpack_from(  
        '>' + label_file_size, label_buf, struct.calcsize('>II'))  
    labels = np.array(labels).astype(np.int64)  
      
    datas_root = '/somePath/mnist_train' #需要修改的路径  
    if not os.path.exists(datas_root):  
        os.mkdir(datas_root)  
      
    for i in range(10):  
        file_name = datas_root + os.sep + str(i)  
        if not os.path.exists(file_name):  
            os.mkdir(file_name)  
      
    for ii in range(numLabels):  
        img = Image.fromarray(datas[ii, 0, 0:28, 0:28])  
        label = labels[ii]  
        file_name = datas_root + os.sep + str(label) + os.sep + \  
            'mnist_train_' + str(ii) + '.png'  
        img.save(file_name)

转换后的目录结构  
![这里写图片描述][SouthEast]  
\#\#\#2、获取KNN训练所需数据  
\#\#\#\#\#为了省事这里直接将图片二值化并转为行向量

def imgtoVect(filename):#filename:图片路径
        f = cv2.imread(filename,0)  
        a,b=f.shape
        Vect=np.array(f).reshape(a*b)#将灰度图转为行向量
        #二值化操作 
        for i in range(a*b): 
            if Vect[i]>127:
                Vect[i]=1
            else:
                Vect[i]=0
        return Vect  
    def createtrainMat(filename):#filename:图片文件夹路径
    
        trainMat=[]
        labels=[]
        root_list=os.listdir(filename)
        for i in root_list:
            img_list=os.listdir(filename+'/'+i)
            for j in img_list:
                labels.append(str(i))#标签和图片上级目录名一致
                trainMat.append(imgtoVect(filename+'/'+i+'/'+j))  
        return trainMat,labels

trainMat和labels大概就是这样

#trainMat
    #每一行代表一张图片的所有特征，每一列代表一类特征
    [[1,0,1,1,...1,0,0],
    [1,1,0,1,...1,1,0],
    [1,0,0,0,...0,0,0],
    	...
    [1,0,1,1,...1,0,0]]
    #labels
    #labels的每一列是trainMat对应行的分类标签
    [0,0,0,...1,1,1...,...9,9,9]

因为训练集太大，每次准备训练数据时间太长，所以我们先将trainMap,labels存储下来，第二次执行的时候可以直接load数据，跳过准备阶段

if os.path.exists('C:/Users/53121/Desktop/Mat.npy'):
    	Mat=np.load('C:/Users/53121/Desktop/Mat.npy')
    	labels=np.load('C:/Users/53121/Desktop/labels.npy')
    else:
    	Mat,labels=createtrainMap("C:/Users/53121/Desktop/mnist_train")
    	np.save('C:/Users/53121/Desktop/Mat.npy', Mat)
    	np.save('C:/Users/53121/Desktop/labels.npy',labels)

\#\#\#3、获取KNN测试所需数据

###### 利用opencv的鼠标事件写数字 ######

3.1、得到手写数字图片

import cv2
    #draw鼠标事件的回掉函数
    def draw(event,x,y,flags,param):
    	global en
    	if event==cv2.EVENT_LBUTTONDOWN:
    		en=True
    	elif event==cv2.EVENT_MOUSEMOVE and flags==cv2.EVENT_LBUTTONDOWN:
    		if en:
    			cv2.circle(img,(x,y),7,255,-1)
    	elif event==cv2.EVENT_LBUTTONUP:
    		en=False
    img=np.zeros((280,280),np.uint8)
    cv2.namedWindow('image') 
    cv2.setMouseCallback('image',draw)
    while(1): 
        cv2.imshow('image',img) 
        if cv2.waitKey(10)&0xFF==27: 
            break
    cv2.destroyAllWindows()

这样就可以得到280\*280的手写体

![这里写图片描述][SouthEast 1]

*  3.2、将图片变为特征矩阵

#首先是尺寸变化，因为mnist图片尺寸为28*28的
    input=cv2.resize(img,(28,28),interpolation=cv2.INTER_CUBIC)
    #转为行向量+二值化
    def imgtoVect2(f):#f:即上面的input
        a,b=f.shape
        Vect=np.array(f).reshape(a*b)
        for i in range(a*b): 
            if Vect[i]>127:
                Vect[i]=1
            else:
                Vect[i]=0
        return Vect

### 4、引入并训练KNN ###

*  训练

from sklearn import neighbors 
    mode = neighbors.KNeighborsClassifier()#创建KNN
    Mat,labels=createtrainMap("C:/Users/53121/Desktop/mnist_train")#获取训练数据
    data = np.array(Mat)
    labels=np.array(labels)
    print '训练mode...'
    mode.fit(data,labels) 
    print 'mode训练完成'

*  分类

print '分类结果：',mode.predict([input])
    print '可能性：' ,mode.predict_proba([input])

### 5、完整代码 ###

# -*- coding: utf-8 -*-
    """
    Created on Sun Nov 05 13:23:21 2017
    
    @author: 53121
    """
    import cv2
    import numpy as np  
    from sklearn import neighbors  
    import os
    from sklearn import tree  
    from sklearn.naive_bayes import GaussianNB
    en=False
    def draw(event,x,y,flags,param):
    	global en
    	if event==cv2.EVENT_LBUTTONDOWN:
    		en=True
    	elif event==cv2.EVENT_MOUSEMOVE and flags==cv2.EVENT_LBUTTONDOWN:
    		if en:
    			cv2.circle(img,(x,y),7,255,-1)
    	elif event==cv2.EVENT_LBUTTONUP:
    		en=False
    def imgtoVect2(f):
        a,b=f.shape
        Vect=np.array(f).reshape(a*b)
        for i in range(a*b): 
            if Vect[i]>127:
                Vect[i]=1
            else:
                Vect[i]=0
        return Vect 
    def imgtoVect(filename):
        f = cv2.imread(filename,0)  
        Vect=np.array(f).reshape(784)
        for i in range(784): 
            if Vect[i]>127:
                Vect[i]=1
            else:
                Vect[i]=0
        return Vect  
    def createtrainMap(filename):
        trainMap=[]
        labels=[]
        root_list=os.listdir(filename)
        for i in root_list:
            img_list=os.listdir(filename+'/'+i)
            for j in img_list:
                labels.append(str(i))
                trainMap.append(imgtoVect(filename+'/'+i+'/'+j))
        return trainMap,labels
    
    method=raw_input('please input method name: ')
    if method=='knn':
        mode = neighbors.KNeighborsClassifier()
    elif method=='tree':
        mode = tree.DecisionTreeClassifier(criterion='gini') 
    elif method=='bayes':
        mode=GaussianNB()
    def classifiy(Mat,labels):        
        data = np.array(Mat)
        labels=np.array(labels)
        print '训练mode...'
        mode.fit(data,labels) 
        print 'mode训练完成'
    img=np.zeros((280,280),np.uint8)
    cv2.namedWindow('image') 
    cv2.setMouseCallback('image',draw)
    if os.path.exists('C:/Users/53121/Desktop/Mat.npy'):
    	Mat=np.load('C:/Users/53121/Desktop/Mat.npy')
    	labels=np.load('C:/Users/53121/Desktop/labels.npy')
    else:
    	Mat,labels=createtrainMap("C:/Users/53121/Desktop/mnist_train")
    	np.save('C:/Users/53121/Desktop/Mat.npy', Mat)
    	np.save('C:/Users/53121/Desktop/labels.npy',labels)
    classifiy(Mat,labels)
    mode.fit(Mat,labels)  
    while(1): 
        cv2.imshow('image',img) 
        if cv2.waitKey(10)&0xFF==27: 
            break
        elif cv2.waitKey(10)&0xFF==115: #按‘s’进行分类
            input=cv2.resize(img,(28,28),interpolation=cv2.INTER_CUBIC)
            input=imgtoVect2(input)         
            print '分类结果：',mode.predict([input])
            print '可能性：' ,mode.predict_proba([input])
        elif cv2.waitKey(10)&0xFF==99:
            img=np.zeros((280,280),np.uint8)
    cv2.destroyAllWindows()

代码里还有决策树和贝叶斯分类方法（写的太乱，可是不想改）  
![这里写图片描述][SouthEast 2]  
其实最后效果不太好，找了个预测对的就贴上来了，哈哈

[http_blog.csdn.net_qq_32166627_article_details_52640730]: http://blog.csdn.net/qq_32166627/article/details/52640730
[SouthEast]: /images/20220606/10a2f28b883e44de95d47b569f2b3aae.png
[SouthEast 1]: /images/20220606/a22b48fa38a24a40a15a2fe21cf0290a.png
[SouthEast 2]: /images/20220606/203fb5c7726d427a8daf3bc07a128df2.png