随着人工智能技术的发展,手写字体识别已成为计算机视觉领域的一个重要研究方向。SVM(支持向量机)是一种常用的机器学习算法,可以用于手写字体识别。
图像链接:https://pan.baidu.com/s/1ydtgCRyyrXCD3VLX1ITnrA?pwd=nqql
提取码:nqql
%%清空环境变量
close all;
clear;
clc;
format compact;
%% 载入训练数据
% 利用uigetfile函数交互式选取训练样本
[FileName,PathName,FilterIndex] = uigetfile( ...
{'*.jpg';'*.bmp'},'请导入训练图片','*.jpg','MultiSelect','on');
if ~FilterIndex
return;
end
num_train = length(FileName);
TrainData = zeros(num_train,16*16);
TrainLabel = zeros(num_train,1);
for k = 1:num_train
pic = imread([PathName,FileName{k}]);
pic = pic_preprocess(pic);
% 将标准化图像按列拉成一个向量并转置,生成50*256的训练样本矩阵
TrainData(k,:) = double(pic(:)');
% 样本标签为样本所对应的数字
TrainLabel(k) = str2double(FileName{k}(4));
end
%% 建立支持向量机
% [bestCVaccuracy,bestc,bestg] = ...
% SVMcgForClass(TrainLabel,TrainData,-8,8,-8,8,10,0.8,0.8,4.5)
% 设置GA相关参数
ga_option.maxgen = 100;
ga_option.sizepop = 20;
ga_option.cbound = [0,100];
ga_option.gbound = [0,100];
ga_option.v = 10;
ga_option.ggap = 0.9;
[bestCVaccuracy,bestc,bestg] = ...
gaSVMcgForClass(TrainLabel,TrainData,ga_option)
gaSVMcgForClass子函数:http://t.csdnimg.cn/02KdD
% 训练
cmd = ['-c ',num2str(bestc),' -g ',num2str(bestg)];
model = svmtrain(TrainLabel, TrainData, cmd);
% 在训练集上查看识别能力
preTrainLabel = svmpredict(TrainLabel, TrainData, model);
%% 载入测试样本
[FileName,PathName,FilterIndex] = uigetfile( ...
{'*.jpg';'*.bmp'},'请导入测试图片','*.bmp','MultiSelect','on');
if ~FilterIndex
return;
end
num_train = length(FileName);
TestData = zeros(num_train,16*16);
TestLabel = zeros(num_train,1);
for k = 1:num_train
pic = imread([PathName,FileName{k}]);
pic = pic_preprocess(pic);
TestData(k,:) = double(pic(:)');
TestLabel(k) = str2double(FileName{k}(4));
end
%% 对测试样本进行分类
preTestLabel = svmpredict(TestLabel, TestData, model);
assignin('base','TestLabel',TestLabel);
assignin('base','preTestLabel',preTestLabel);
TestLabel'
preTestLabel'
子函数pic_preprocess
function pic_preprocess = pic_preprocess(pic)
% 图片预处理子函数
% 图像反色处理
pic = 255-pic;
% 设定阈值,将反色图像转成二值图像
pic = im2bw(pic,0.4);
% 查找数字上所有像素点的行标y和列标x
[y,x] = find(pic == 1);
% 截取包含完整数字的最小区域
pic_preprocess = pic(min(y):max(y), min(x):max(x));
% 将截取的包含完整数字的最小区域图像转成16*16的标准化图像
pic_preprocess = imresize(pic_preprocess,[16,16]);
源自:《Matlab神经网络43个案例分析》。