Python handwriting recognition system based on KNN algorithm

Tag:float   gdi    Add     two value image    shape    ext    Environment    color    presentation    

This paper mainly uses K-nearest neighbor classifier to realize handwriting recognition system, training data set about 2000 samples, each number has about 200 samples, each sample is saved in a TXT file, the handwriting image itself is a 32x32 two value image, as shown in:  Test code for handwritten numeral recognition system: From numpy import *import operatorfrom os import listdir   #inX     The data to be detected #dataset    DataSet #labels    result set #k      length to compare def  Classify0 (inx, dataset, labels, k):    datasetsize = dataset.shape[0]                  #计算有多少行     # tile (inx,  (datasetsize,1)) Generate a matrix of corresponding InX dimensions for easy     Diffmat = tile (inx,  (datasetsize,1))  - dataSet    sqdiffmat = diffmat**2                          #差求平方     sqdistances = sqdiffmat.sum (Axis=1)               # axis=0,  represents a column  axis=1,  represents a row.     distances = sqdistances**0.5  &NBsp                 #开方     sorteddistindicies =  Distances.argsort ()         #argsort () sort, subscript     classcount={}    For i  in range (k):        voteilabel = labels[sorteddistindicies[i]]    #通过下标索引分类         #  by constructing a dictionary, recording classification frequency         Classcount[voteilabel]  = classcount.get (voteilabel,0)  + 1        #  sort fields by value (from big to small)         sortedclasscount = sorted (Classcount.items (), Key=lambda classcount:classcount [1], reverse=true)     return sortedclasscount[0][0]  #手写字体识别 # First, we need to format the image as a vector, #   Converts a 32x32 binary image matrix through the Img2vector () function to the 1x1024 vector: def img2vector (filename):    returnvect =  zeros (1,1024)     fr = open (filename)     For i in range (+): &NBSP, #图片矩阵为32 *32        Linestr = fr.readline ()     #数据量大, so use readline        For j in range (+):            returnvect[0,32*i+j]& Nbsp;= int (Linestr[j])     return returnvect  #手写字体识别def  handwritingclasstest ():     hwlabels = []    Trainingfilelist = listdir (R ' trainingdigits ')             #指定文件夹     M = len (trainingfilelist)         & nbsp                         #获取文件夹个数     trainingmat  = zeros (m,1024)                             &N BSP, #构造m个1024比较矩阵     For i in range (m):        filenamestr =  trainingfilelist[i]          #获取文件名 &NBSp       Filestr = filenamestr.split ('. ') [0]        #按点把文件名字分割         Classnumstr = int (Filestr.split ('_') [0] )     #按下划线把文件名字分割         Hwlabels.append (CLASSNUMSTR)           & nbsp     #实际值添加保存         Trainingmat[i,:] = img2vector (R ' trainingdigits/%s '   % FILENAMESTR)         Testfilelist = listdir (' testdigits ')         #测试数据         errorcount = 0.0        Mtest = len ( testfilelist)     For i in range (mtest): #同上, processing test data         FILENAMESTR  = testFileList[i]        filestr = filenamestr.split ('. ') [0]      #take  off .txt        Classnumstr = int (filestr.split (‘_‘) [0])   &NBSp     Vectorundertest = img2vector (R ' testdigits/%s '  % filenamestr)         Classifierresult = classify0 (vectorundertest, trainingmat, hwlabels, 3)         print  ("Calculated value: %d,  actual value:  %d"  %  (classifierresult,  CLASSNUMSTR)         if  (CLASSIFIERRESULT != CLASSNUMSTR):  errorcount  += 1.0    print  ("\ n error occurrences:  %d"  % errorcount)     print  ("\ N Error Rate:  %f " %  (Errorcount/float (mtest))) handwritingclasstest () result:  calculated value: 9, actual value: 9 Calculated value: 9, actual value: 9 Calculated value: 9, actual value: 9 Calculated value: 9, actual value: 9 Calculated value: 9, actual value: 9 Calculated value: 9, actual value: 9  error occurred: 10  error Rate: 0.010571  can see the KNN algorithm for memory consumption is very large (i 12G), Chinese environment recognition I can't imagine.

Python's handwriting recognition system based on KNN algorithm