Matlab: Classification

-
(German) This post describes a Matlab function which classifies images dependent on shape color and so on to differentiate between images like bananas and apples.

Mode of operation:

In this function different attributes of images are calculated in values to classify images and seperate them in 2D or 3D. Classification methodes are 'color', 'color2' and 'colorhex' for different color classifications and 'brightness' for the brightness value 'shape', 'std', 'mean', 'cov' and 'sub' are different shape conditions. This function can be downloaded on MathWorks. Following figure shows the classification of apples, bananas and black berries. 


It is seen in three dimensions it is clearly classified. Most of the mathematical algortihm are based on magnitude gradient maps to whether calculate shape or substructures. Following figure shows the different input images and their gradient map.
In source code give below there is an example how to use it. MLClassification(image,class_1,class_2,class_3) classifies the image according to class 1-3 with 3 output arguments. Iteratively over all images this is done and then plotted. With parzen window approach a Gaussian cloud is fitted around the values with a standard deviation of 2 to border the different image groups. The function Gaussian cloud is given at the very bottom of the source code. 
To determine where an image belongs the function IdentifyClass(ax1, ax2, ax3, ax4, varargin) is used. It calculates the minimal distance to the Gaussian cloud surface of the different groups. Because mathematically elliptic bodies have no analytical solutions the function [~, dist] = PointOnSurface(XData, YData, ZData, arg1, arg2, arg3); calculates it numerically which can be downloaded here. 

Source code:

tic;
%% ____________________Plotting___________________%%
class_1 = 'color';
class_2 = 'shape';
class_3 = 'sub';
% Processing Apple
tresh = 0.35;
cd(dir_apple);
for i=1:1:length(files_apple)
    img_apple{i} = im2double(imread(names_apple{i}));
    image = img_apple{i};
    [arg1_apple(i), arg2_apple(i), arg3_apple(i)] = MLClassification(image,class_1,class_2,class_3);
end

% Processing Apple Green
cd(dir_others);
for i=1:1:length(files_others)
    img_others{i} = im2double(imread(names_others{i}));
    image = img_others{i};
    [arg1_others(i), arg2_others(i), arg3_others(i)] = MLClassification(image,class_1,class_2,class_3);
end

% Processing Banana
cd(dir_banana);
for i=1:1:length(files_banana)
    img_banana{i} = im2double(imread(names_banana{i}));
    image = img_banana{i};
    [arg1_banana(i), arg2_banana(i), arg3_banana(i)] = MLClassification(image,class_1,class_2,class_3);
end

col_1 = 'darkred';
col_2 = 'yellow';
col_3 = 'black';
fig(2) = figure(2);clf(2);
dev = 1.2;
ax1 = subplot(2,2,1);hold on;box on;%set(gca,'xtick',[]);set(gca,'ytick',[]);
plot(arg1_apple,arg2_apple,'.','Markersize',12,'Color',clrs(col_1));
plot(arg1_banana,arg2_banana,'.','Markersize',12,'Color',clrs(col_2));
plot(arg1_others,arg2_others,'.','Markersize',12,'Color',clrs(col_3));
GaussCloud(arg1_apple,arg2_apple,'Color',clrs(col_1),'dev',dev);
GaussCloud(arg1_banana,arg2_banana,'Color',clrs(col_2),'dev',dev);
GaussCloud(arg1_others,arg2_others,'Color',clrs(col_3),'dev',dev);
% LinearDiscriment(arg2_apple,arg1_apple,arg2_banana,arg1_banana);
xlabel(class_1);ylabel(class_2);

ax2 = subplot(2,2,2);hold on;box on;%set(gca,'xtick',[]);set(gca,'ytick',[]);
plot(arg1_apple,arg3_apple,'.','Markersize',12,'Color',clrs(col_1));
plot(arg1_banana,arg3_banana,'.','Markersize',12,'Color',clrs(col_2));
plot(arg1_others,arg3_others,'.','Markersize',12,'Color',clrs(col_3));
GaussCloud(arg1_apple,arg3_apple,'Color',clrs(col_1),'dev',dev);
GaussCloud(arg1_banana,arg3_banana,'Color',clrs(col_2),'dev',dev);
GaussCloud(arg1_others,arg3_others,'Color',clrs(col_3),'dev',dev);
xlabel(class_1);ylabel(class_3);

ax3 = subplot(2,2,3);hold on;box on;%set(gca,'xtick',[]);set(gca,'ytick',[]);
plot(arg3_apple,arg2_apple,'.','Markersize',12,'Color',clrs(col_1));
plot(arg3_banana,arg2_banana,'.','Markersize',12,'Color',clrs(col_2));
plot(arg3_others,arg2_others,'.','Markersize',12,'Color',clrs(col_3));
GaussCloud(arg3_apple,arg2_apple,'Color',clrs(col_1),'dev',dev);
GaussCloud(arg3_banana,arg2_banana,'Color',clrs(col_2),'dev',dev);
GaussCloud(arg3_others,arg2_others,'Color',clrs(col_3),'dev',dev);
xlabel(class_3);ylabel(class_2)

ax4 = subplot(2,2,4);hold on;box on;%set(gca,'xtick',[]);set(gca,'ytick',[]);set(gca,'ztick',[]);
plot3(arg1_apple,arg2_apple,arg3_apple,'.','Markersize',12,'Color',clrs(col_1));
plot3(arg1_banana,arg2_banana,arg3_banana,'.','Markersize',12,'Color',clrs(col_2));
plot3(arg1_others,arg2_others,arg3_others,'.','Markersize',12,'Color',clrs(col_3));
GaussCloud(arg1_apple,arg2_apple,arg3_apple,'Color',clrs(col_1),'dev',dev);
GaussCloud(arg1_banana,arg2_banana,arg3_banana,'Color',clrs(col_2),'dev',dev);
GaussCloud(arg1_others,arg2_others,arg3_others,'Color',clrs(col_3),'dev',dev);
xlabel(class_1);ylabel(class_2);zlabel(class_3);view(15,15);axis tight;

time_2 = toc;

% __________Load Image for Classification__________$
choice = questdlg('Load image for classification?', ...
    'Question DLG', ...
    'Yes','No','No');
switch choice
    case 'Yes'
        use_own = 1;
    case 'No'
        use_own = 0;
end

if use_own == 1
    id = IdentifyClass(ax1 ,ax2, ax3, ax4,'args',class_1,class_2,class_3);
end
%% ____________________Identify Class_____________________%%
function varargout = IdentifyClass(ax1, ax2, ax3, ax4, varargin)
% Identify to which class the loaded image belongs to with the distance
% function
ax_dist = get(ax4);

varargout = cell(length(varargin),1);
logic = strcmp(varargin,'args');
if any(logic)~= 0
    args = find(logic,1);
    class_1 = varargin{args+1};
    class_2 = varargin{args+2};
    class_3 = varargin{args+3};
else
    class_1 = 'color';
    class_2 = 'shape';
    class_3 = 'std';
end

[filename, pathname] = uigetfile({'*.jpg';'*.bmp';'*.gif';'*.*'},'File Selector');
cd(pathname);

img_test = im2double(imread(filename));
col_test = 'blue';
[arg1_test, arg2_test, arg3_test] = MLClassification(img_test,class_1,class_2,class_3);

XData1 = ax_dist.Children(5).XData;YData1 = ax_dist.Children(5).YData;ZData1 = ax_dist.Children(5).ZData;
XData2 = ax_dist.Children(3).XData;YData2 = ax_dist.Children(3).YData;ZData2 = ax_dist.Children(3).ZData;
XData3 = ax_dist.Children(1).XData;YData3 = ax_dist.Children(1).YData;ZData3 = ax_dist.Children(1).ZData;

[~, dist_1] = PointOnSurface(XData1, YData1, ZData1, arg1_test, arg2_test, arg3_test);
[~, dist_2] = PointOnSurface(XData2, YData2, ZData2, arg1_test, arg2_test, arg3_test);
[~, dist_3] = PointOnSurface(XData3, YData3, ZData3, arg1_test, arg2_test, arg3_test);

dist = [dist_1 dist_2 dist_3];

varargout{1} = dist;
fprintf('The loaded image belongs to class %d', I);

plot(ax1,arg1_test,arg2_test,'.','Markersize',12,'Color',clrs(col_test));
plot(ax2,arg1_test,arg3_test,'.','Markersize',12,'Color',clrs(col_test));
plot(ax3,arg3_test,arg2_test,'.','Markersize',12,'Color',clrs(col_test));
plot3(ax4,arg1_test,arg2_test,arg3_test,'.','Markersize',12,'Color',clrs(col_test));
%
% plot(ax1,surf_1(1),surf_1(2),'x','Markersize',12,'Color',clrs(col_test));
% plot(ax2,surf_1(1),surf_1(3),'x','Markersize',12,'Color',clrs(col_test));
% plot(ax3,surf_1(3),surf_1(2),'x','Markersize',12,'Color',clrs(col_test));
% plot3(ax4,surf_1(1),surf_1(2),surf_1(3),'x','Markersize',12,'Color',clrs(col_test));
   
end  

%% ________________Gauss Cloud_________________________%%
function [middle, distances, stds] = GaussCloud(x,y,varargin)

if isstring(varargin{1})
    ex = 0;
    z = zeros(numel(x),1);
else
    z = varargin{1};
    ex = 1;
end
% Color defintion
logic_col = strcmp(varargin,'Color');
if any(logic_col)~= 0
    pos_col = find(logic_col,1);
    col = varargin{pos_col+1};
else
    col = [0 0 0];
end
% Size
logic_col = strcmp(varargin,'dev');
if any(logic_col)~= 0
    pos_col = find(logic_col,1);
    dev = varargin{pos_col+1};
else
    dev = 1;
end
%Markersize definition
logic_size = strcmp(varargin,'Markersize');
if any(logic_size)~= 0
    pos_mark = find(logic_size,1);
    mark_size = varargin{pos_mark+1};
else
    mark_size = 12;
end
% Transparency definition
logic_alpha = strcmp(varargin,'Alpha');
if any(logic_alpha)~= 0
    pos_alpha = find(logic_alpha,1);
    alpha = varargin{pos_alpha+1};
else
    alpha = 0.25;
end

% The Function

mean_x = mean(x);
mean_y = mean(y);
mean_z = mean(z);
middle = [mean_x,mean_y,mean_z];

std_x = std(x);
std_y = std(y);
std_z = std(z);
stds = [std_x,std_y,std_z];

vec_x = mean_x-x;mean_dist_x = mean(vec_x);
vec_y = mean_y-y;mean_dist_y = mean(vec_y);
vec_z = mean_z-z;mean_dist_z = mean(vec_z);
distances = [mean_dist_x, mean_dist_y, mean_dist_z];
   
% generate sphere
[X,Y,Z] = sphere(100);
X = X.*(2*std_x*dev);
Y = Y.*(2*std_y*dev);
Z = Z.*(2*std_z*dev);

X = X+mean_x;
Y = Y+mean_y;
Z = Z+mean_z;

hold on;
if ex == 1
    plot3(mean_x,mean_y,mean_z,'x','Markersize',mark_size,'Color',col);
    surf(X,Y,Z,'FaceAlpha',alpha,'FaceColor',col,'EdgeColor','none');
else
    plot(mean_x,mean_y,'x','Markersize',mark_size,'Color',col);
    surf(X,Y,Z,'FaceAlpha',alpha,'FaceColor',col,'EdgeColor','none');
end

end

Kommentare

Kommentar veröffentlichen

Beliebte Posts aus diesem Blog

Matlab: 3D Coordinate System Rotations with Vectors

-
Bild
This post describes how to switch/transform between corrdinate systems in Matlab and how this can be represented. There are 3 functions that are crucial here: grid3D(startVec, stopVec, varargin) ... Creates an orthogonal grid in 3D depending on start and end point (diagonal of a cube) and draws ticks, labels and MinorGrids. Rotations can be performed here to show how to transform between Cartesian corrdinate systems. point3D(startVec, stopVec, varargin) ... Is a function to create 3D vectors based on start and end points. It includes options like naming the points, displaying them as arrows, as well as settings for color and thickness. arrow3D(startVec, stopVec, varargin) ... This function creates an arrow in 3D that can be scaled by the options. The functions including example can be downloaded here: Matlab-Functions/CartesianCorrdinateTransformation at main · PhysIngo/Matlab-Functions (github.com) First of all an example where the 1 coordinate system with 2 vectors is rotated 2 times...
Mehr anzeigen

Matlab: Cone/Arrow in 3D

-
Cone/Arrow in 3D To plot a cone or arrow in 3D you can use this function: function arrow3D(startVec, stopVec, varargin) The startVec is the beginning of the 3D cone and the stopVec the end of the cone. Ypi can use the optional arguments as 'Color','Thickness','Length' to customize your cone. Source code : function arrow3D(startVec,stopVec,varargin) % Input arguments: % startVec ... input vector of the beginning of the point in [x,y,z] % stopVec ... input vector of the ending of the point in [x,y,z] % % Optional input argument (varargin): % 'Color',color ... define the color of the arrow/point % 'Thickness',thick ... thickness of the arrow % 'Length',length ... Length of the arrow % 'Axes',ax ... Input axes     if isempty(varargin)         varargin{1} = '';     end          [x] = startVec(1);      [y] = startVec(2);      [z] = startVec(3);      x2 = stopVec(1);   ...
Mehr anzeigen

Matlab: Points and vectors in 3D-plots

-
Bild
Points and Vectors in 3D-Plots (German) This matlabfunction creates vectors and points for the previous post in the 3D cartesian coordinate system. function punkt3D(x1,y1,z1,x2,y2,z2,sign,name,color,dx,dy,dz,true) Therefor "startVec" are the initial values of the vector and "stopVec" the final values. You need the function ( Cone/Arrow ) and  ( Grid3D ). The optional arguments are shown below. For example you can use it as: figure;hold on;axis off; grid3D([0,0,0], [1,1,1], 'Amount', [4,4,4], 'Axes', gca, 'MinorGrid','Arrow','AxTicks',[{['']},{['']},{['']}],...     'AxLabels',["";"";""]); P = [0.5,0.5,1]; Q = [1,0.5,1]; point3D([0,0,0],P,'Marker','arrow','Label','P(0.5|0.5|1)','Color',[0.7 0.5 0]); point3D([0,0,0],Q,'Marker','arrow','Label','Q(1|0.5|1)','Color',[0.0 0.4 0.74],'D...
Mehr anzeigen