www.gusucode.com > IPCV_Eval_Kit_R2019a_0ce6858工具箱matlab程序源码 > IPCV_Eval_Kit_R2019a_0ce6858/code/demo_files/I5_04_4_myHOGDigitClassification_Tree.m

    %% HOG (Histogram of Oriented Gradient) 摿挜検 偲
%  寛掕栘乮2暘栘乯 傪巊偭偨丄庤彂偒悢帤偺幆暿
clear;clc;close all;imtool close all

% 僩儗乕僯儞僌夋憸乮101枃x10暥帤庬乯偲僥僗僩夋憸乮12枃x10暥帤庬乯傊偺愨懳僷僗傪愝掕
pathData = [toolboxdir('vision'), '\visiondata\digits']
trainSet = imageSet([pathData,'\synthetic'  ], 'recursive');
testSet  = imageSet([pathData,'\handwritten'], 'recursive');

%% 慡僩儗乕僯儞僌梡夋憸椺偺昞帵
figure;montage([trainSet.ImageLocation], 'Size', [26 40]);

%% 慡僥僗僩夋憸傪儌儞僞乕僕儏昞帵 (12枃 x 10暥帤庬丗奺庤彂偒悢帤傪擣幆)
figure;montage([testSet(:).ImageLocation], 'Size', [10,12]);

%% 4x4偺僙儖僒僀僘傪巊梡 (324師尦儀僋僩儖)
cellSize = [4 4];
hogFeatureSize = 324;                   % length(hog_4x4)

%% [暘椶栘偺峔抸]丗fitctree傪巊梡
% 10暥帤暘偺trainingFeatures 傪奿擺偡傞攝楍傪偁傜偐偠傔嶌惢
trainingFeatures  = zeros(10*101,hogFeatureSize, 'single');
trainingLabels    = zeros(10*101,1);

% HOG摿挜検傪拪弌
for digit = 0:9   % 暥帤'0'乣'9'
  for i = 1:101         % 奺悢帤偛偲偵101枃偺僩儗乕僯儞僌梡夋憸
    img = read(trainSet(digit+1), i);  %僩儗乕僯儞僌夋憸偺撉崬傒       trainSet()偼丄1偐傜巒傑傞偺偱丄+1
    img = im2bw(img,graythresh(img));   % 擇抣壔
             
    trainingFeatures((digit)*101+i,:) = extractHOGFeatures(img,'CellSize',cellSize);
    trainingLabels((digit)*101+i)     = digit;
  end
end
% 暘椶栘偺妛廗
treeModel = fitctree(trainingFeatures, trainingLabels, 'MaxNumSplits',4)

%% 暘椶僣儕乕價儏乕傾乕偱丄惗惉偟偨僣儕乕傪昞帵
%    x斣崋偼丄摿挜検偺斣崋
view(treeModel, 'mode','graph');

%% [幆暿] 嶌惉偟偨暘椶婍偱庤彂偒悢帤(120枃)傪幆暿\帵丗predict()
Ir = zeros([16,16,3,120], 'uint8');      % 寢壥傪奿擺偡傞攝楍
cntTrue = 0;
for digit = 0:9   % 
  for i = 1:12         % 奺悢帤偛偲偵12枃偺庤彂偒暥帤
    img = read(testSet(digit+1), i);    % testSet()偼丄1偐傜巒傑傞偺偱丄+1
    BW = im2bw(img,graythresh(img));    % 2抣壔

    testFeatures = extractHOGFeatures(BW,'CellSize',cellSize);
    predictedNum = predict(treeModel, testFeatures);           % testFeature 傪攝楍偵偟偰丄偁偲偱傑偲傔偰敾掕傕壜
    
    if predictedNum == digit    %惓偟偄幆暿偼惵怓丄岆擣幆偼愒怓
      Ir(:,:,:,digit*12+i) = insertText(img,[6 4],num2str(predictedNum),'FontSize',9,'TextColor','blue','BoxOpacity',0.4);
      cntTrue = cntTrue+1;
    else
      Ir(:,:,:,digit*12+i) = insertText(img,[6 4],num2str(predictedNum),'FontSize',9,'TextColor','red','BoxOpacity',0.4); 
    end 

  end
end
% 寢壥偺昞帵
figure;montage(Ir, 'Size', [10,12]); title(['Correct Prediction: ' num2str(cntTrue)]);

%%


%% 傾儞僒儞僽儖妛廗 (ClassificationBaggedEnsemble 僋儔僗偑曉傞)
%   僶僊儞僌寛掕栘 => 儔儞僟儉僼僅儗僗僩 傾儖僑儕僘儉傪巊梡
template = templateTree('MaxNumSplits', 20);
bagModel = fitcensemble(...
	trainingFeatures, trainingLabels, ...
	'Method', 'Bag', ...
	'NumLearningCycles', 30, 'Learners', template);

%% bagModel.Trained偼丄 CompactClassificationTree僋儔僗偺僙儖攝楍
%       Children : 僲乕僪斣崋弴偵丄偦偺僲乕僪偑偮側偑偭偰偄傞2偮偺僲乕僪斣崋丅廔抂僲乕僪偵娭偟偰偼0 0 偑擖傞
%       CutPredictor : 奺僲乕僪偱偺暘婒偵梡偄傞摿挜検斣崋
%       CutPoint : 暘婒偺鑷抣 (廔抂僲乕僪偵娭偟偰偼丄NaN偑擖傞)
%       NodeClass : 奺僲乕僪偱偺暘椶偝傟偨僋儔僗
view(bagModel.Trained{3})
view(bagModel.Trained{3}, 'mode','graph')




%% 暘椶妛廗婍傾僾儕働乕僔儑儞偺巊梡
%   僶僊儞僌寛掕栘偼丄儔儞僟儉僼僅儗僗僩
%   峔憿懱偺儊儞僶乕偲偟偰丄ClassificationEnsemble (CompactClassificationEnsemble僋儔僗)  偑曉傞
%   奺ClassificationEnsemble.Trained{k}偑丄CompactClassificationTree僋儔僗偺僆僽僕僃僋僩
dataTable = table(trainingFeatures, trainingLabels, 'VariableNames',{'features', 'label'});     % 1x324 偺摿挜儀僋僩儖 + 儔儀儖   偑丄200峴
openvar('dataTable');
classificationLearner

% 僐儞僷僋僩儌僨儖偺僄僋僗億乕僩丗trainedClassifier
view(trainedClassifier.ClassificationEnsemble.Trained{1}, 'mode','graph')

%% [幆暿] 嶌惉偟偨暘椶婍偱庤彂偒悢帤(120枃)傪幆暿\帵丗trainedClassifier.predictFcn()
Ir = zeros([16,16,3,120], 'uint8');      % 寢壥傪奿擺偡傞攝楍
cntTrue = 0;
for digit = 0:9   % 
  for i = 1:12         % 奺悢帤偛偲偵12枃偺庤彂偒暥帤
    img = read(testSet(digit+1), i);    % testSet()偼丄1偐傜巒傑傞偺偱丄+1
    BW = im2bw(img,graythresh(img));    % 2抣壔

    features = extractHOGFeatures(BW,'CellSize',cellSize);
		predictedNum = trainedClassifier.predictFcn(table(features));    % testFeature 傪攝楍偵偟偰丄偁偲偱傑偲傔偰敾掕傕壜
 
    if predictedNum == digit    %惓偟偄幆暿偼惵怓丄岆擣幆偼愒怓
      Ir(:,:,:,digit*12+i) = insertText(img,[6 4],num2str(predictedNum),'FontSize',9,'TextColor','blue','BoxOpacity',0.4);
      cntTrue = cntTrue+1;
    else
      Ir(:,:,:,digit*12+i) = insertText(img,[6 4],num2str(predictedNum),'FontSize',9,'TextColor','red','BoxOpacity',0.4); 
    end 

  end
end
% 寢壥偺昞帵
figure;montage(Ir, 'Size', [10,12]); title(['Correct Prediction: ' num2str(cntTrue)]);




%% Copyright 2013-2014 The MathWorks, Inc.
% 夋憸僨乕僞僙僢僩
% 僩儗乕僯儞僌夋憸丗insertText娭悢偱帺摦嶌惉 (廃埻偵暿偺悢帤桳傝)
% 僥僗僩夋憸丗庤彂偒偺夋憸傪巊梡