一、课题研究的意义
汉字已有数千年的历史,是中华民族文化的重要结晶,闪烁着中国人民智慧的光芒。同时也是世界上使用人数最多和数量最多的文字之一。现如今,汉字印刷材料的数量大大增加,一些专业单位所接触的印刷材料更是浩如烟海,信息量均是爆炸性增长。然而,汉字是非字母化、非拼音化的文字,因此,如何将汉字快速高效地输入计算机,是信息处理的一个关键问题,也是关系到计算机技术能否在我国真正普及的关键问题,更是传播与弘扬中华民族悠久历史文化的关键问题。而且随着劳动力价格的升高,利用人工方法进行汉字输入也将面临经济效益的挑战。因此,对于大量已有的文档资料,汉字自动识别输入就成为了最佳的选择。因此,汉字识别技术也越来越受到人们的重视。汉字识别是一门多学科综合的研究课题,它不仅与人工智能的研究有关,而且与数字信号处理、图像处理、信息论、计算机科学、几何学、统计学、语言学、生物学、模糊数学、决策论等都有着千丝万缕的联系。一方面各学科的发展给它的研究提供了工具;另一方面,它的研究与发展也必将促进各学科的发展。因而有着重要的实用价值和理论意义。
二、算法流程
![25eff3593b6fdf5e2baa5f5fc9ff2bc2.png](https://img-blog.csdnimg.cn/img_convert/25eff3593b6fdf5e2baa5f5fc9ff2bc2.png)
三、识别过程
![191a1d790dfc38723e3f01567266a162.png](https://img-blog.csdnimg.cn/img_convert/191a1d790dfc38723e3f01567266a162.png)
![be58833f44dbfb558156c2dd3c935877.png](https://img-blog.csdnimg.cn/img_convert/be58833f44dbfb558156c2dd3c935877.png)
![d74fe9c6bf04b62a76188c7089e39807.png](https://img-blog.csdnimg.cn/img_convert/d74fe9c6bf04b62a76188c7089e39807.png)
![9e513b2765e97ecc0848b530e6aea8df.png](https://img-blog.csdnimg.cn/img_convert/9e513b2765e97ecc0848b530e6aea8df.png)
![d726ec3008579b2d6277478dcc79c14a.png](https://img-blog.csdnimg.cn/img_convert/d726ec3008579b2d6277478dcc79c14a.png)
通过对标准图像识别学习训练,然后对输入乱序图像(见图2-8)进行识别,过程中转灰度(见图2-9),均值滤波(见图2-10),二值化(见图2-11),识别结果(见图2-12)。
![4e4c1fa7a6ca15c5152be50fba554010.png](https://img-blog.csdnimg.cn/img_convert/4e4c1fa7a6ca15c5152be50fba554010.png)
![70f0bdcb3deda4a07c7777ac3796d489.png](https://img-blog.csdnimg.cn/img_convert/70f0bdcb3deda4a07c7777ac3796d489.png)
![ee23280dfd0ea900906268c319730293.png](https://img-blog.csdnimg.cn/img_convert/ee23280dfd0ea900906268c319730293.png)
![0bdc3c764c4b1b5402556db8a5bb519d.png](https://img-blog.csdnimg.cn/img_convert/0bdc3c764c4b1b5402556db8a5bb519d.png)
![2ca451304a05eddab036e1897f868141.png](https://img-blog.csdnimg.cn/img_convert/2ca451304a05eddab036e1897f868141.png)
由以上实验结果可以看出,当输入标准图像,经过转灰度,均值滤波,二值化,识别输出。然后,输入乱序图像在经过转灰度,均值滤波,二值化,特征提取,匹配识别后,系统能够将乱序图像中的图像识别输出。说明系统基本实现预先设想的功能,能够在学习标准图像识别后建立标准库,并根据标准字库与以后输入的图像进行匹配识别输出。但由于系统比较简单,实现的功能也比较单一,要想实现较完备功能还需进一步完善。
四、部分参考源码
%乱序图像识别:
load bp;
G=imread('');%读取
I=rgb2gray(G);
%--------转灰度图象
B1=filter2(fspecial('average',3),I)/255;
%以[3,3]为模板均值滤波
%--------均值滤波
d=im2bw(B1,;
%--------二值
k1=1;k2=1;s=sum(d');j=2;a=1;c=1
[m, n]=size(d');
while s(j)==m
j=j+1;
end
k1=j;
while s(j)~=m && j<=n-1
j=j+1;
end
k2=j-1;
d=d((k1:k2),:);
%--------行分割
[m,n]=size(d);
k1=1;k2=1;s=sum(d);j=2;a=1;c=1
for a=1:16
while s(j)==m
j=j+1;
end
k1=j;
while s(j)~=m && j<=n-1
j=j+1;
end
k2=j-1;
%--------列分割
if c==1
b1=d(:,(k1:k2));
imwrite(b1,'');
t1=tezhengtiqu(b1);
P_test = [t1'];
y = sim(net,P_test);
%用训练出来的神经网络计算数据的第%P_test行
word=jieguoxianshi(y)
end
%--------第1个字符
if c==2
b2=d(:,(k1:k2));
imwrite(b2,'');
t2=tezhengtiqu(b2);
%用十三点特征提取法提取特征
P_test = [t2];
y = sim(net,P_test');
%用训练出来的神经网络计算数据的
%第P_test行
word=jieguoxianshi(y)
end
%--------第2个字符
if c==3
b3=d(:,(k1:k2));
imwrite(b3,'');
t3=tezhengtiqu(b3);
P_test = [t3];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第3个字符
if c==4
b4=d(:,(k1:k2));
imwrite(b4,'');
t4=tezhengtiqu(b4);
P_test = [t4];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第4个字符
if c==5
b5=d(:,(k1:k2));
imwrite(b5,'');
t5=tezhengtiqu(b5);
P_test = [t5];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第5个字符
if c==6
b6=d(:,(k1:k2));
imwrite(b6,'');
t6=tezhengtiqu(b6);
P_test = [t6];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第6个字符
if c==7
b7=d(:,(k1:k2));
imwrite(b7,'');
t7=tezhengtiqu(b7);
P_test = [t7];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第7个字符
if c==8
b8=d(:,(k1:k2));
imwrite(b8,'');
t8=tezhengtiqu(b8);
P_test = [t8];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第8个字符
if c==9
b9=d(:,(k1:k2));
imwrite(b9,'');
t9=tezhengtiqu(b9);
P_test = [t9];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第9个字符
if c==10
b10=d(:,(k1:k2));
imwrite(b10,'');
t10=tezhengtiqu(b10);
P_test = [t10];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第10个字符
if c==11
b11=d(:,(k1:k2));
imwrite(b11,'');
t11=tezhengtiqu(b11);
P_test = [t11];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第11个字符
if c==12
b12=d(:,(k1:k2));
imwrite(b12,'');
t12=tezhengtiqu(b12);
P_test = [t12];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第12个字符
if c==13
b13=d(:,(k1:k2));
imwrite(b13,'');
t13=tezhengtiqu(b13);
P_test = [t13];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第13个字符
if c==14
b14=d(:,(k1:k2));
imwrite(b14,'');
t14=tezhengtiqu(b14);
P_test = [t14];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第14个字符
if c==15
b15=d(:,(k1:k2));
imwrite(b15,'');
t15=tezhengtiqu(b15);
P_test = [t15];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第15个字符
if c==16
b16=d(:,(k1:k2));
imwrite(b16,'');
t16=tezhengtiqu(b16);
P_test = [t16];
y = sim(net,P_test');
word=jieguoxianshi(y)
end
%--------第16个字符
c=c+1;
end
subplot(2,8,1),imshow('');
subplot(2,8,2),imshow('');
subplot(2,8,3),imshow('');
subplot(2,8,4),imshow('');
subplot(2,8,5),imshow('');
subplot(2,8,6),imshow('');
subplot(2,8,7),imshow('');
subplot(2,8,8),imshow('');
subplot(2,8,9),imshow('');
subplot(2,8,10),imshow('');
subplot(2,8,11),imshow('');
subplot(2,8,12),imshow('');
subplot(2,8,13),imshow('');
subplot(2,8,14),imshow('');
subplot(2,8,15),imshow('');
subplot(2,8,16),imshow('');
%标准图像识别:
load fbp;
G=imread('');%读取
I=rgb2gray(G);
%--------转灰度图象
B1=filter2(fspecial('average',3),I)/255;
%--------均值滤波
imshow(d);
k1=1;k2=1;s=sum(d');j=2;a=1;c=1
[m,n]=size(d');
while s(j)==m
j=j+1;
end
k1=j;
while s(j)~=m && j<=n-1
j=j+1;
end
k2=j-1;
d=d((k1:k2),:);
%--------行分割
[m,n]=size(d);
k1=1;k2=1;s=sum(d);j=2;a=1;c=1
for a=1:16
while s(j)==m
j=j+1;
end
k1=j;
while s(j)~=m && j<=n-1
j=j+1;
end
k2=j-1;
%--------列分割
if c==1
b1=d(:,(k1:k2));
imwrite(b1,'');
t1=tezhengtiqu(b1);
P_test = [t1'];
y = sim(net,P_test);
word=fbmjieguoxianshi(y)
end
%--------第1个字符
if c==2
b2=d(:,(k1:k2));
imwrite(b2,'');
t2=tezhengtiqu(b2);
P_test = [t2];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第2个字符
if c==3
b3=d(:,(k1:k2));
imwrite(b3,'');
t3=tezhengtiqu(b3);
P_test = [t3];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第3个字符
if c==4
b4=d(:,(k1:k2));
imwrite(b4,'');
t4=tezhengtiqu(b4);
P_test = [t4];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第4个字符
if c==5
b5=d(:,(k1:k2));
imwrite(b5,'');
t5=tezhengtiqu(b5);
P_test = [t5];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第5个字符
if c==6
b6=d(:,(k1:k2));
imwrite(b6,'');
t6=tezhengtiqu(b6);
P_test = [t6];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第6个字符
if c==7
b7=d(:,(k1:k2));
imwrite(b7,'');
t7=tezhengtiqu(b7);
P_test = [t7];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第7个字符
if c==8
b8=d(:,(k1:k2));
imwrite(b8,'');
t8=tezhengtiqu(b8);
P_test = [t8];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第8个字符
if c==9
b9=d(:,(k1:k2));
imwrite(b9,'');
t9=tezhengtiqu(b9);
P_test = [t9];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第9个字符
if c==10
b10=d(:,(k1:k2));
imwrite(b10,'');
t10=tezhengtiqu(b10);
P_test = [t10];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第10个字符
if c==11
b11=d(:,(k1:k2));
imwrite(b11,'');
t11=tezhengtiqu(b11);
P_test = [t11];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第11个字符
if c==12
b12=d(:,(k1:k2));
imwrite(b12,'');
t12=tezhengtiqu(b12);
P_test = [t12];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第12个字符
if c==13
b13=d(:,(k1:k2));
imwrite(b13,'');
t13=tezhengtiqu(b13);
P_test = [t13];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第13个字符
if c==14
b14=d(:,(k1:k2));
imwrite(b14,'');
t14=tezhengtiqu(b14);
P_test = [t14];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第14个字符
if c==15
b15=d(:,(k1:k2));
imwrite(b15,'');
t15=tezhengtiqu(b15);
P_test = [t15];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第15个字符
if c==16
b16=d(:,(k1:k2));
imwrite(b16,'');
t16=tezhengtiqu(b16);
P_test = [t16];
y = sim(net,P_test');
word=fbmjieguoxianshi(y)
end
%--------第16个字符
c=c+1;
end
subplot(2,8,1),imshow('');
subplot(2,8,2),imshow('');
subplot(2,8,3),imshow('');
subplot(2,8,4),imshow('');
subplot(2,8,5),imshow('');
subplot(2,8,6),imshow('');
subplot(2,8,7),imshow('');
subplot(2,8,8),imshow('');
subplot(2,8,9),imshow('');
subplot(2,8,10),imshow('');
subplot(2,8,11),imshow('');
subplot(2,8,12),imshow('');
subplot(2,8,13),imshow('');
subplot(2,8,14),imshow('');
subplot(2,8,15),imshow('');
subplot(2,8,16),imshow('');
%标准图像识别结果显示:
function word=jieguoxianshi(y)
%显示结果的函数
y=[round(y(1)),round(y(2)),round(y(3)),
round(y(4)),round(y(5)),round(y(6)),
round(y(7)),round(y(8)),round(y(9)),
round(y(10)),round(y(11)),round(y(12)),
round(y(13)),round(y(14)),round(y(15)),
round(y(16))];
[C,I]=max(y);
if I==16
word='匣';
elseif I==15
word='囡';
elseif I==14
word='图';
elseif I==13
word='国';
elseif I==12
word='盅';
elseif I==11
word='匡';
elseif I==10
word='盏';
elseif I==9
word='固';
elseif I==8
word='监';
elseif I==7
word='盂';
elseif I==6
word='盒';
elseif I==5
word='团';
elseif I==4
word='盈';
elseif I==3
word='団';
elseif I==2
word='盔';
elseif I==1
word='因';
end
%乱序图像识别结果显示
function word=jieguoxianshi(y)
%显示结果的函数
y=[round(y(1)),round(y(2)),round(y(3)),
round(y(4))];
if y==[0,0,0,0]
word='匣';
elseif y==[0,0,0,1,]
word='囡';
elseif y==[0,0,1,0]
word='图';
elseif y==[0,0,1,1]
word='国';
elseif y==[0,1,0,0]
word='盅';
elseif y==[0,1,0,1]
word='匡';
elseif y==[0,1,1,0]
word='盏';
elseif y==[0,1,1,1]
word='固';
elseif y==[1,0,0,0]
word='监';
elseif y==[1,0,0,1]
word='盂';
elseif y==[1,0,1,0]
word='盒';
elseif y==[1,0,1,1]
word='团';
elseif y==[1,1,0,0]
word='盈';
elseif y==[1,1,0,1]
word='団';
elseif y==[1,1,1,0]
word='盔';
elseif y==[1,1,1,1]
word='因';
end
%特征提取:
function PN=moshishibie(d)
%特征提取函数
%通过13点特征提取法提取特征
%d为已处理图象
[m,n]=size(d);
k1=1;
for i=1:m/4
for j=1:n/2
if d(i,j)==1
k1=k1+1;
end
end
end
k2=1;
for i=1:round(m/4)
for j=round(n/2):round(n)
if d(i,j)==1
k2=k2+1;
end
end
end
k3=1;
for i=round(m/4):m/2
for j=1:n/2
if d(i,j)==1
k3=k3+1;
end
end
end
k4=1;
for i=round(m/4):m/2
for j=round(n/2):n
if d(i,j)==1
k4=k4+1;
end
end
end
k5=1;
for i=round(m/2):round(m*3/4)
for j=1:round(n/2)
if d(i,j)==1
k5=k5+1;
end
end
end
k6=1;
for i=round(m/2):round(m*3/4)
for j=round(n/2):n
if d(i,j)==1
k6=k6+1;
end
end
end
k7=1;
for i=round(m*3/4):m
for j=1:round(n/2)
if d(i,j)==1
k7=k7+1;
end
end
end
k8=1;
for i=round(m*3/4):m
for j=round(n/2):n
if d(i,j)==1
k8=k8+1;
end
end
end
k9=k3+k4;
k10=k5+k6;
k11=k1+k3+k5+k7;
k12=k2+k4+k6+k8;
k13=k11+k12;
k=[k1,k2,k3,k4,k5,k6,k7,k8,k9,k10,k11,k12,13];
[PN,minp,maxp] = premnmx(k); %归一化
附 录2
Neural Network object: %神经网络的对象;
architecture: %结构;
numInputs: 1
numLayers: 3
biasConnect: [1; 1; 1]
inputConnect: [1; 0; 0]
layerConnect: [0 0 0; 1 0 0; 0 1 0]
outputConnect: [0 0 1]
targetConnect: [0 0 1]
numOutputs: 1 (read-only)
numTargets: 1 (read-only)
numInputDelays: 0 (read-only)
numLayerDelays: 0 (read-only)
subobject structures: %子对象结构
inputs: {1x1 cell} of inputs
layers: {3x1 cell} of layers
outputs: {1x3 cell} containing 1 output
targets: {1x3 cell} containing 1 target
biases: {3x1 cell} containing 3 biases
inputWeights: {3x1 cell} containing 1 input weight
layerWeights: {3x3 cell} containing 2 layer weights
functions: %功能
adaptFcn: 'trains'
initFcn: 'initlay'
performFcn: 'mse'
trainFcn: 'trainlm'
parameters: %参数
adaptParam: .passes
initParam: (none)
performParam: (none)
trainParam: .epochs, .goal, .max_fail, .mem_reduc,
.min_grad, .mu, .mu_dec, .mu_inc,
.mu_max, .show, .time
weight and bias values: %权重和偏置值
IW: {3x1 cell} containing 1 input weight matrix
LW: {3x3 cell} containing 2 layer weight matrices
b: {3x1 cell} containing 3 bias vectors
other: %其他
userdata: (user stuff)