#python#实现最近邻识别手写字

最近邻算法比较简单，但是样本集多了之后比较的次数会非常多，所以我这里选择的是用均值算每一个类的模板类，通过和模板类对比来识别手写字

#单模板分类器，加快分类速度
import cv2 as cv
import numpy as np
import time


def getdata(path):
    img = cv.imread(path,0)#参数0为读成灰度图片
    return cv.resize(img,(15,15))#修规图片大小

def change(m):
    """把一个数据矩阵变成一行"""
    M = []
    for i in range(m.shape[1]):
        temp = m[:][i]
        M=M+list(temp)
    return M

#K近邻分类器
def KNNClassfy(X, S, K):
    #一行为一个样本，一列为一个像素
    #X，S均为numpy矩阵
    x = X.shape[0]
    y = S.shape[0]
    dist =np.ones((x,y))
    for i in range(x):
        for j in range(y):
            a = X[i]#识别样本
            b = S[j]#训练样本
            dist[i,j] = np.sqrt(np.sum(np.square(a - b)))
    T = np.argsort(dist, axis=1)[:,0:K]
    num_id = []
    for i in range(x):
        result = -np.array([-1]*K)
        for j in range(K):
            result[j] = T[i,j]
        _,n=sorted([(np.sum(result==i),i) for i in set(result)])[-1]
        num_id.append(n)
    return num_id

#通用路径
train_path = "D:\\MyWorkPlace\\pattern recognition\\images4000"
test_path = "D:\\MyWorkPlace\\pattern recognition\\images_test1000"


#训练集载入
print("开始载入数据集！",end = " ")
my_train = []
for i in range(10):
    tmp = []
    for j in range(400):  
        img = getdata(train_path+f'\\{i}_{j}.bmp')
        tmp.append(change(img))
    tmp_arr = np.array(tmp)
    print("\r训练集已载入：{:0.2f}%".format((i*400+j+1)/40),end=" ")
    my_train.append(np.mean(tmp_arr, axis=0))
S = np.array(my_train)

#测试集载入
my_test = []
for i in range(10):
    for j in range(401,501):  
        img = getdata(test_path+f'\\{i}_{j}.bmp')
        my_test.append(change(img))
    print("\r测试集已载入：{:0.2f}%".format((i*100+j-400)*100/1000),end=" ")
print("\r数据集已载入{:0.2f}%,载入完毕！".format((i*100+j-400)*100/1000))
X = np.array(my_test)
#K取1即最近邻分类器
time_start = time.time() #开始计时
result = KNNClassfy(X,S,1)#分类器输出标签
#创建正确标签true_result
true_result = []
for i in range(10):
    true_result = true_result+[i]*100
count = 0
for i in range(len(result)):
    if result[i]==true_result[i]:
        count = count+1
    else:
        print("预测为%2s   实际为%2s"%(result[i],true_result[i]))
print("正确率：",count/len(result))
time_end = time.time() 
print("用时{:0.2f}s".format(time_end-time_start))