smiuser/ofm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

classification code

Browse Source
This commit is contained in:
moonanjum26
2018-08-23 11:57:56 +05:30
parent ab3036cec2
commit 2261a6e825
3 changed files with 287 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/albina nirupa/classificationcode.m Normal file
View File

@@ -0,0 +1,27 @@
%%
mkdir('E:\Experiments\odroid\A2\A2a\cells\rbc');
mkdir('E:\Experiments\odroid\A2\A2a\cells\cluster');
mkdir('E:\Experiments\odroid\A2\A2a\cells\wbc');
mkdir('E:\Experiments\odroid\A2\A2a\cells\parasite');
%%
%Classifier Function obtained after training the classifier with the
%generated training data set and the feature table
% There are 4 different catergories of classification
X = table2array(varfun(@double, CellFeature_table_A(:,trainedClassifiercells.PredictorNames)));
cellclass = predict(trainedClassifiercells, X);
for i = 1:length(cellclass)
image = imread(strcat('E:\Experiments\odroid\A2\A2a\cells\A1 (',num2str(i),').jpg'));
if(cellclass(i)== 'rbc')
imwrite(image,strcat('E:\Experiments\odroid\A2\A2a\cells\rbc\',num2str(i),'.jpg'),'jpeg');
else if (cellclass(i)== 'wbc')
imwrite(image,strcat('E:\Experiments\odroid\A2\A2a\cells\wbc\',num2str(i),'.jpg'),'jpeg');
else if (cellclass(i)== 'cluster')
imwrite(image,strcat('E:\Experiments\odroid\A2\A2a\cells\cluster\',num2str(i),'.jpg'),'jpeg');
else if (cellclass(i)== 'parasite')
imwrite(image,strcat('E:\Experiments\odroid\A2\A2a\cells\parasite\',num2str(i),'.jpg'),'jpeg');
end
end
end
end
end

37
src/albina nirupa/createMask.m Normal file
View File

@@ -0,0 +1,37 @@
function [BW,maskedRGBImage] = createMask(RGB,Imax,Imin)
%createMask Threshold RGB image using auto-generated code from colorThresholder app.
% [BW,MASKEDRGBIMAGE] = createMask(RGB) thresholds image RGB using
% auto-generated code from the colorThresholder App. The colorspace and
% minimum/maximum values for each channel of the colorspace were set in the
% App and result in a binary mask BW and a composite image maskedRGBImage,
% which shows the original RGB image values under the mask BW.
% Auto-generated by colorThresholder app on 12-Dec-2015
%------------------------------------------------------
% Convert RGB image to chosen color space
I = RGB;
% Define thresholds for channel 1 based on histogram settings
channel1Min = Imin;%132.000;
channel1Max = Imax;%182.000;
% Define thresholds for channel 2 based on histogram settings
channel2Min = channel1Min;%131.000;
channel2Max = channel1Max;%177.000;
% Define thresholds for channel 3 based on histogram settings
channel3Min = channel1Min;%130.000;
channel3Max = channel1Max;%180.000;
% Create mask based on chosen histogram thresholds
BW = (I(:,:,1) >= channel1Min ) & (I(:,:,1) <= channel1Max) & ...
(I(:,:,2) >= channel2Min ) & (I(:,:,2) <= channel2Max) & ...
(I(:,:,3) >= channel3Min ) & (I(:,:,3) <= channel3Max);
% Initialize output masked image based on input image.
maskedRGBImage = RGB;
% Set background pixels where BW is false to zero.
maskedRGBImage(repmat(~BW,[1 1 3])) = 0;

223
src/albina nirupa/whole_blood_segment_odroid.m Normal file
View File

@@ -0,0 +1,223 @@
%%
%The code creates a feature table for the different types of cells and
%is used to create a initial training dataset for the classifier
%%
clc ;
close all;
clear all;
%%
mkdir('E:\Experiments\odroid\S\S1\gate1');
mkdir('E:\Experiments\odroid\S\S1\gate2');
mkdir('E:\Experiments\odroid\S\S1\gate3');
mkdir('E:\Experiments\odroid\S\S1\gate4');
mkdir('E:\Experiments\odroid\S\S1\P');
%%
% To create a circular mask to eliminate the spokes channels and perform
% processing of cells in the central imaging region
% Create a logical image of a circle with specified
% diameter, center, and image size.
% First create the image.
imageSizeX = 640; %X pixel size of the image
imageSizeY = 480; %Y pixel size of the image
[columnsInImage rowsInImage] = meshgrid(1:imageSizeX, 1:imageSizeY);
% Next create the circle in the image.
centerX = 300; %center X pixel of the circle
centerY = 240; %center y pixel of the circle
radius = 220; %Radius of the circle
circlePixels = (rowsInImage - centerY).^2 ...
+ (columnsInImage - centerX).^2 <= radius.^2; % creates the circular mask
%% Generate Background
%a = rgb2gray(imread(strcat('E:\Experiments\odroid\step1\1 (',num2str(V),').jpg')));
%imshow(a);
%%% average ten frames to generate background.
bg = 0; V=1;
count = 500 ;% set number of frame to be averaged
N = 1; % start frame number
for i = N:N+count
bg = bg +double(rgb2gray((imread(strcat('E:\Experiments\odroid\S\S1\1 (',num2str(V+i),').jpg')))));
end
bg = uint8(bg /count); % Final Background generated.
imshow(uint8(circlePixels).*bg);
%%
tic
r =0;b=0;
count = 0;
maxglcm = 0;
Cont =0;Corr=0;Homo=0;I_mean=0;I_std=0;I_cir=0;I_area =0;stain=0;
ccount = 0;
C_Percentstained =0;StainMaxlength=0;StainMinlength=0;StainSolid=0; StainNumobj=0;
Features_table = [];F=[];
CellFeature_table = [];cell=[];
% Feature_table = table('VariableNames',{'Area', 'ConvexArea', 'Eccentricity', 'EquivDiameter', 'EulerNumber', 'Extent', 'FilledArea', 'MajorAxisLength', 'MinorAxisLength', 'Orientation', 'Perimeter', 'Solidity'});
for V =1:29999
disp(V);
Shape_Features =[]; Texture_Features1 = [];
% The if condition is used to refresh the background for every 2000 frames
% This helps eliminate any debris/ struck cells in the ROI
if mod(V,2000) == 0
%if V <19999-30
% bg1 = bg;
bg =0;
count = 300 ;% set number of frame to be averaged
N = 1; % start frame number
for i = N:N+count
bg = bg +double(rgb2gray((imread(strcat('E:\Experiments\odroid\S\S1\1 (',num2str(V+i),').jpg')))));
end
bg = uint8(bg /count); % Final Background generated.
% bg = (bg+bg1)/2.0
end
% The following section performs the segmentation based on histogram values
CurrFrame = (rgb2gray((imread(strcat('E:\Experiments\odroid\S\S1\1 (',num2str(V),').jpg')))));
Sub =double(double(CurrFrame.*uint8(circlePixels))-double(bg.*uint8(circlePixels)));
%imshow(Sub);
Submin = min(Sub(:));
Submax = max(Sub(:));
AdjBGSub = uint8( (Sub - Submin)/(Submax-Submin) * 255);
Ia = AdjBGSub;
[A, B]=size(Ia);
I = medfilt2(Ia); % To smoothen the image
I = adapthisteq(I);
I1 = medfilt2(I);
Irgb = cat(3, I1, I1, I1);
meanIntensityValue(V) = mean2(I1); % Finds the mean of the intensities of the image pixels
stdIntensityValue = std2(I1); % Finds the standard deviation of the intensities of the image pixels
% Selection of max and minimum of intensities for the thresholding
%This multipication factor can be varied when you are optimizing the
%thresholding
Imax = meanIntensityValue(V)+stdIntensityValue*4; % mean+ 4*standard deviation
Imin = meanIntensityValue(V)-stdIntensityValue*4; % mean- 4*standard deviation
% Further processing of the thresholded image
Mask = createMask(Irgb,Imax,Imin);%figure;imshow(Mask);
MaskInv = ~Mask;
Maskopen =bwareaopen(MaskInv,150);%figure;imshow(Maskfinal);
% Maskdil = imdilate(Maskopen, [se90 se0]);%figure;imshow(Maskdil);
Maskdil = imclose(Maskopen, strel('disk',5));
Maskfill = imfill(Maskdil, 'holes');%figure;imshow(Maskfill);
Maskclose = Maskfill;
% Maskclose = imclose(Maskfill, strel('disk',5));
Maskclear = imclearborder(Maskclose, 4);%figure;imshow(Maskclear);
Maskfinal = Maskclear;
% Filter image based on image properties.
Maskfinal = bwpropfilt(Maskfinal, 'Area', [350 + eps(350), Inf]); % Area greater than 350 pixels
Maskfinal = bwpropfilt(Maskfinal, 'Solidity', [0.6 + eps(0.6), Inf]); % 1 is completely solid region
Maskfinal = bwpropfilt(Maskfinal, 'EulerNumber', [4.94065646e-324 + eps(4.94065646e-324), Inf]);
%Extract properties of all the cells in the thresholded image
Maskproperties = regionprops(Maskfinal, {'Area', 'ConvexArea', 'Eccentricity', 'EquivDiameter', 'EulerNumber', 'Extent', 'FilledArea', 'MajorAxisLength', 'MinorAxisLength', 'Orientation', 'Perimeter', 'Solidity'});
% Creates a feature table of all the above listed properties for all the
% segmented cells/regions in the image
Shape_Features = struct2table(Maskproperties);
%Plotting of 200 images to see the performance of segmetation operations
if V<200
h= figure;
subplot(2,3,1);
subimage(Mask);
title('Mask');
subplot(2,3,2);
subimage(Maskfill);
title('Maskfill');
subplot(2,3,3);
subimage(Maskdil);
title('Maskdil');
subplot(2,3,4);
subimage(uint8(MaskInv).*I1);
title('Maskclose');
% subplot(1,2,1);
subplot(2,3,5);
subimage(I1);
title('I1');
% subplot(1,2,2);
subplot(2,3,6);
subimage(uint8(Maskfinal).*I1);
title(' Maskfinal');
%saveas(h,strcat('E:\Experiments\odroid\S\S1\gate4\FrameNumber','-',num2str(V)),'jpg');
close(h);
% figure;imshow(uint8(Maskfinal).*I);
% figure;imshow(I);
end
% Generation of the traning images for the classification program
CComp = bwconncomp(Maskfinal);
Areas = regionprops(CComp,'Area');
Centroids1=regionprops(CComp,'Centroid');
Perimeters=regionprops(CComp,'Perimeter');
Texture_Features=[];
for i = 1:CComp.NumObjects
Cent = Centroids1(i).Centroid;
HighX = round(Cent(1))+20;
if HighX>B
HighX =B;
end
HighY = round(Cent(2))+20;
if HighY>A
HighY =A;
end
LowX = round(Cent(1))-19;
if LowX<=0
LowX =1;
end
LowY = round(Cent(2))-19;
if LowY<=0
LowY =1;
end
ccount = ccount +1;%,num2str(V),'.avi'
Icrop=Ia(LowY:HighY,LowX:HighX);
Icrop1=I(LowY:HighY,LowX:HighX);
Maskcrop=Maskfinal(LowY:HighY,LowX:HighX);
imwrite(Icrop,strcat('E:\Experiments\odroid\S\S1\gate3\FrameNumber','-',num2str(V),'_',num2str(ccount),'.jpg'));
% Additional texture features for the feature table
glcm = graycomatrix(Icrop1);%gray level covariance matrix
maxglcm(ccount)=max(max(glcm)); %Feature 1
stats(ccount) = graycoprops(glcm,{'Contrast','Correlation','homogeneity'});
Texture_Features = struct2table(stats(ccount));
% Cont(ccount)= stats(ccount).Contrast; %Feature 2
% Corr(ccount) = stats(ccount).Correlation; %Feature 3
% Homo(ccount)= stats(ccount).Homogeneity;%Feature 4
% Ent(ccount)= stats(ccount).Entropy;
I_mean(ccount) = mean2(Icrop1); %Feature 5
I_std(ccount)=std2(Icrop1); %Feature 6
I_cir(ccount) = ((Shape_Features.Perimeter(i))^ 2)/ (4 * pi * Shape_Features.Area(i)); %Feature 7
% Percentage of stained area
BWthresh = im2bw(Icrop,0.15);
BWthreshtemp =~(BWthresh).*Maskcrop;
CC1 =bwconncomp(imcomplement(BWthresh).*Maskcrop);
MaxLength = regionprops(CC1,'MajorAxisLength');
MinLength = regionprops(CC1,'MinorAxisLength');
Solid = regionprops(CC1,'Solidity');
% imwrite(~BWthresh,strcat('E:\Experiments\odroid\A2\A2a\gate2\FrameNumber','-',num2str(V),'_',num2str(ccount),'.jpg'));
Areas_stained = regionprops(CC1,'Area');
StainNumobj(ccount) = CC1.NumObjects;
if CC1.NumObjects ==0
C_Percentstained(ccount) = 0;
StainMaxlength(ccount) = 0;StainMinlength(ccount) = 0;
StainSolid(ccount) = 0; StainSolid(ccount) = 0;
else
if (max(struct2array(MaxLength))/max(struct2array(MinLength)))<2
F = [F; max(struct2array(MaxLength))/max(struct2array(MinLength))];
% imwrite(Icrop,strcat('E:\Experiments\odroid\A2\A2a\P\FrameNumber','-',num2str(V),'_',num2str(ccount),'.jpg'));
end
C_Percentstained(ccount) = sum(struct2array(Areas_stained))/max((Shape_Features.Area(i)))*100;
StainMaxlength(ccount) = max(struct2array(MaxLength));
StainMinlength(ccount) = max(struct2array(MinLength));
StainSolid(ccount) = max(struct2array(Solid));
% Stainproperties = regionprops(BWthreshtemp, {'Area', 'ConvexArea', 'Eccentricity', 'EquivDiameter', 'EulerNumber', 'Extent', 'FilledArea', 'MajorAxisLength', 'MinorAxisLength', 'Orientation', 'Perimeter', 'Solidity'});
% Parasite_Features = struct2table(Stainproperties);
end
Texture_Features1 = [Texture_Features; Texture_Features1];
end
if CComp.NumObjects>0
Features = [Shape_Features Texture_Features1]; %Variable Addition
Features_table = [Features_table; Features]; % Table update
end
cell = table(I_mean', I_std', I_cir', maxglcm', C_Percentstained',StainMaxlength',StainMinlength',StainSolid', StainNumobj');
cell.Properties.VariableNames = {'Mean' 'Std' 'Circularity' 'MaxGLCM' 'Percentage_Stain' 'StainMaxlength' 'StainMinlength' 'StainSolid' 'StainNumobj'};
CellFeature_table = [Features_table cell]; % Table update
end
toc