K-Nearest Neighbors (KNN) for recognizing handwritten digits

Beijing Institute of Technology | Ming-Jian Li

The following Python code is a companion code for the course on Artificial Intelligence and Simulation Science. It functions to convert an image of a handwritten digit, and use a KNN model to recognize the number. The dataset is in data.zip (click to download) .

x
1
# convert image into txt
2
import os
3
from PIL import Image
4

5
def imgtotext(imgfile,txtfile,size=(32,32)):
6
    # convert image to txt file, with size 32*32
7
    image_file = Image.open(imgfile)
8
    image_file = image_file.resize(size,Image.LANCZOS)
9
    image_file=image_file.convert('L')
10
    width,height = image_file.size
11
    f =open(txtfile,'w')
12
    ascii_char = '10'
13
    for i in range(height):
14
        pix_char='';
15
        for j in range(width):
16
            pixel = image_file.getpixel((j,i))
17
            pix_char+=ascii_char[int(pixel/128)]
18
        pix_char+='\n'
19
        f.write(pix_char)
20
    f.close()
21

22
imgtotext(r'test.jpg', r'test.txt')
23

24
from os import listdir
25
from numpy import *
26
import numpy as np
27
import operator
28

29
def KNN(test_data,train_data,train_label,k):
30
    # rows of dataset
31
    dataSetSize = train_data.shape[0]
32
    # expand input points to matrix, get distance
33
    all_distances = np.sqrt(np.sum(np.square(tile(test_data,(dataSetSize,1))-train_data),axis=1))
34
    print("all distances: ",all_distances)
35
    # sort distance
36
    sort_distance_index = all_distances.argsort()
37
    # choose k points with min distances
38
    classCount = {}
39
    for i in range(k):
40
        voteIlabel = train_label[sort_distance_index[i]]
41
        classCount[voteIlabel] = classCount.get(voteIlabel,0)+1
42
    sortedClassCount = sorted(classCount.items(), key = operator.itemgetter(1), reverse = True)
43
    return sortedClassCount[0][0]
44

45
# turn txt into vector 32x32 -> 1x1024
46
def img2vector(filename):
47
    returnVect = []
48
    fr = open(filename)
49
    for i in range(32):
50
        lineStr = fr.readline()
51
        for j in range(32):
52
            returnVect.append(int(lineStr[j]))
53
    return returnVect
54

55
# get label from file name
56
def classnumCut(fileName):
57
    # ref file name: 0_3.txt
58
    fileStr = fileName.split('.')[0]
59
    classNumStr = int(fileStr.split('_')[0])
60
    return classNumStr
61

62
# training set
63
def trainingDataSet():
64
    train_label = []
65
    trainingFileList = listdir('data/trainingDigits')
66
    m = len(trainingFileList)
67
    train_data = zeros((m,1024))
68
    # get label
69
    for i in range(m):
70
        fileNameStr = trainingFileList[i]
71
        train_label.append(classnumCut(fileNameStr))
72
        train_data[i,:] = img2vector('data/trainingDigits/%s' % fileNameStr)
73
    return train_label,train_data
74

75
# test number
76
Nearest_Neighbor_number = 3
77
train_label,train_data = trainingDataSet()
78
test_data = img2vector('test.txt')
79
# predict
80
classifierResult = KNN(test_data, train_data, train_label, Nearest_Neighbor_number)
81
print ("I guess the number is: ", classifierResult)

The following figure is a handwritten digit:

The prediction by KNN is:

I guess the number is: 6