opencv+KNN实现手写简单数字识别

学长要求做一个可视化的与数据分析的程序，就想到了这个，用了最简单的方法，其实识别效率不是很高，基本上书写要与训练数据比较相似才能识别出来，正在想办法去优化，先把目前的成果分享出来，本人是用python+opencv做的。

draw.py

相当于一个写字板，将写完字后的图片保存转为32*32

#encoding:utf-8
import cv2
import numpy as np #mouse callback function
from knn import *
ix,iy=-1,-1
#创建图像与窗口并将窗口与回调函数绑定
def in_img():
    for i in range(512):
        img[i,:]=255
    cv2.namedWindow('image')
    cv2.setMouseCallback('image',draw_circle)
    while(1):
        cv2.imshow('image',img)
        if cv2.waitKey(20)& 0xFF == ord('q'):
            cv2.imwrite( '1.jpg',img)
            break
    cv2.destroyAllWindows()
def draw_circle(event,x,y,flags,param):
    global ix,iy,drawing
    if event==cv2.EVENT_LBUTTONDOWN:
        drawing=True
        ix,iy=x,y
    elif event==cv2.EVENT_MOUSEMOVE and flags==cv2.EVENT_FLAG_LBUTTON:
        if drawing==True:
            cv2.circle(img,(x,y),30,(0,0,0),-1)
    elif event==cv2.EVENT_LBUTTONUP:
        drawing==False
def read_image():
    img1 = cv2.imread('1.jpg', cv2.IMREAD_GRAYSCALE)
    res=cv2.resize(img1,(32,32),interpolation=cv2.INTER_CUBIC)
    pic=[]
    for i in range(32):
        for j in range(32):
            if res[i][j]<=200:
                res[i][j]=1
            else:
                res[i][j]=0
            pic.append(int(res[i][j]))
    hwLabels,trainingMat = trainingDataSet()
    classifierResult = classify(pic, trainingMat, hwLabels, 3)
    print classifierResult
def main():
    global img
    img=np.zeros((512,512,3),np.uint8)
    in_img()
    read_image()
if __name__=="__main__":
    main()

knn.py

利用KNN算法找到离测试数据最近的10个点（程序中可以调），然后在这10个数据中选取出现次数最多的，即为测试数据代表的数据。

#encoding: utf-8
from numpy import *
import operator
import time
from os import listdir
def classify(inputPoint,dataSet,labels,k):
    dataSetSize = dataSet.shape[0]	 #已知分类的数据集（训练集）的行数
    #先tile函数将输入点拓展成与训练集相同维数的矩阵，再计算欧氏距离
    diffMat = tile(inputPoint,(dataSetSize,1))-dataSet  #样本与训练集的差值矩阵
    sqDiffMat = diffMat ** 2					#差值矩阵平方
    sqDistances = sqDiffMat.sum(axis=1)		 #计算每一行上元素的和
    distances = sqDistances ** 0.5			  #开方得到欧拉距离矩阵
    sortedDistIndicies = distances.argsort()	#按distances中元素进行升序排序后得到的对应下标的列表
    #选择距离最小的k个点
    classCount = {}
    for i in range(k):
        voteIlabel = labels[ sortedDistIndicies[i] ]
        classCount[voteIlabel] = classCount.get(voteIlabel,0)+1
    #按classCount字典的第2个元素（即类别出现的次数）从大到小排序
    sortedClassCount = sorted(classCount.items(), key = operator.itemgetter(1), reverse = True)
    return sortedClassCount[0][0]
def img2vector(filename):
    returnVect = []
    fr = open(filename)
    for i in range(32):
        lineStr = fr.readline()
        for j in range(32):
            returnVect.append(int(lineStr[j]))
    return returnVect
def classnumCut(fileName):
    fileStr = fileName.split('.')[0]
    classNumStr = int(fileStr.split('_')[0])
    return classNumStr
#构建训练集数据向量，及对应分类标签向量
def trainingDataSet():
    hwLabels = []
    trainingFileList = listdir('trainingDigits1')		   #获取目录内容
    m = len(trainingFileList)
    trainingMat = zeros((m,1024))						  #m维向量的训练集
    for i in range(m):
        fileNameStr = trainingFileList[i]
        hwLabels.append(classnumCut(fileNameStr))
        trainingMat[i,:] = img2vector('trainingDigits1/%s' % fileNameStr)
        #print type(trainingMat)
    return hwLabels,trainingMat

我个人还写了个数据，可以自己建立一些测试数据，对于书写有着特殊风格的人可以尝试，当然如果你非常感兴趣的话，还可以用这个程序建立关于英文字母的测试数据，然后用上面的KNN识别英文字母

train_data.py

#encoding:utf-8
import cv2
import numpy as np #mouse callback function
from knn import *
ix,iy=-1,-1
#创建图像与窗口并将窗口与回调函数绑定
def in_img():
    for i in range(512):
        img[i,:]=255
    cv2.namedWindow('image')
    cv2.setMouseCallback('image',draw_circle)
    while(1):
        cv2.imshow('image',img)
        if cv2.waitKey(20)& 0xFF == ord('q'):
            cv2.imwrite( '1.jpg',img)
            break
    cv2.destroyAllWindows()
def draw_circle(event,x,y,flags,param):
    global ix,iy,drawing
    if event==cv2.EVENT_LBUTTONDOWN:
        drawing=True
        ix,iy=x,y
    elif event==cv2.EVENT_MOUSEMOVE and flags==cv2.EVENT_FLAG_LBUTTON:
        if drawing==True:
            cv2.circle(img,(x,y),20,(0,0,0),-1)
    elif event==cv2.EVENT_LBUTTONUP:
        drawing==False
def read_image():
    img1 = cv2.imread('1.jpg', cv2.IMREAD_GRAYSCALE)
    res=cv2.resize(img1,(32,32),interpolation=cv2.INTER_CUBIC)
    pic=[]
    for i in range(32):
        for j in range(32):
            if res[i][j]<=200:
                res[i][j]=1
            else:
                res[i][j]=0
            pic.append(int(res[i][j]))
    f=open('trainingDigits/6_9.txt','w')
    for i in range(32):
        for j in range(32):
            f.write(str(res[i][j]))
        f.write("\n")
    f.close()
def main():
    global img
    img=np.zeros((512,512,3),np.uint8)
    in_img()
    read_image()
if __name__=="__main__":
    main()

结语：个人有一个想法，因为原先的测试数据太多，再加上KNN本身只属于一种算法，并不是一种可建立的模型，每一次识别的时间花费非常长，利用KNN+Adaboost算法尝试尝试，感兴趣的话可以尝试一下

数据来源：http://pan.baidu.com/s/1eQIM4Lk?errno=0&errmsg=Auth%20Login%20Sucess&stoken=9002846861134dfc8b364739cf183e78edcb825ba47e18b1fcfca477d2863a584d122304c8a8739cdcea6bdc3729b78c16da28d4fde8764a0e6b93faf67a6af4d40a3b562d36&bduss=7a7eb001d8d08e2603aaf1732e36c0aa6d77a46d41a907ceb00444543799daedca40105c7c29c9ad32852f293a6013324a71513422ba2cb7cb6b48b3c7679e28c47126a05d0266522fa40883397d046b45914a76628db9b3faefe0e76c5ef6ca53477ffd4a80f84c0d070ff26f672d6c1f00d2aad9b4d034924b1cea44720b551f522bd7160652a38e6b4a851640396f9e93d02424b8674f7c55035c9d5fb8499098a7624e3aa82470a2df78f8585583a1a00f163046b736c36f4ab03ab08269923fcdbb4bfa&ssnerror=0

最后贴一张效果图：