Heelveel content?

content/conclusion.tex

@@ -1 +1,3 @@
-\section{conclusion}
+\section{conclusion}
+
+In general face recognition systems are classified by a single receiver operating characteristic but analyzing a system performance by the single ROCs as in this paper shows a more detailed recognition performance. Although the number of subjects was pretty small to get significant results there were some easy to spot differences between subjects and between different pictures. The failure to capture rate was only a small $1.2\%$. This FCR is made up by 8 pictures who failed almost all face captures and there is a trend in these pictures. All these pictures have a very bad lighting condition, the clothing has far more illumination than the face. But other conditions certainly influence the matching score. Longer hair has a negative influence on the matching scores this has a possible correlation with the gender of the subject. But their are no female subjects with short hair or male subjects with long hair to test this correlation. Unfortunately is this subject group to small to test the Verilook software. But with a larger group of subjects it can be a good procedure to determine where recognition software makes mistakes.

content/discussion.tex

@@ -0,0 +1,2 @@
+\section{Discussion}
+There is a relative high number of genuine scores with the score 0. When looking to these pictures there is not always a clear difference which could be an explanation for this score. Expanding the subject group might resolve this problem and give the possibility to draw conclusions on why this scores are 0. As stated in the results the long hair has one of the bigger influences. But it is also very difficult to separate different hairstyles. Hairstyles change overtime and even different hair types like curly or strait hair can have a big influence. Using different datasets can result in other interesting results. Where a uneven background could have such a large influence that the type of hair is not even significant anymore.

content/introduction.tex

@@ -14,4 +14,4 @@
 % 
 % Here we have the typical use of a "T" for an initial drop letter
 % and "HIS" in caps to complete the first word.
-\IEEEPARstart{F}{or} the assignment of Introduction to Biometrics an analysis needs to be done on Verilook and the Face REcognition Grand Challenge (FRGC) database. Where Verilook is state of the art face recognition software and the FRGC database consists of 50,000 recordings. The Verilook system will generate matching scores between two pictures from the FRGC database. The verification performance is usually abstracted from a single ROC. It is possible though to test this performance on a individual ROC\rq{}s. Which gives the possibility to find cases and/or individuals for which the verification performance drops.
+\IEEEPARstart{F}{or} the assignment of Introduction to Biometrics an analysis needs to be done on Verilook and the Face Recognition Grand Challenge (FRGC) database. Where Verilook is state of the art face recognition software and the FRGC database consists of 50,000 recordings. The Verilook system will generate matching scores between two pictures from the FRGC database. The verification performance is usually abstracted from a single Receiver operating characteristic (ROC). It is possible though to test this performance on a individual ROCs. Which gives the possibility to find cases and/or individuals for which the verification performance drops.

content/method.tex

@@ -6,7 +6,7 @@ Since it is interesting to know why some individuals have better recognizing per
 
 To get a reliable estimating of the true-match rate there should be enough pictures per individual. A minimum of 20 pictures per individual is chosen. 
 
-Test was performed with 31 individuals. From every individual are exact 20 pictures in the dataset. So in total are there 620 different pictures which will be analyzed.  
+Test was performed with 31 individuals. From every individual are exact 20 pictures in the dataset. So in total are there 620 different pictures which will be analyzed.  All images have a relative high resolution of 1704~by~2272 pixels.
 
 \subsection{Get the score matrix from Verilook}
 Verilook is face recognition software. Verilook can compare two pictures from individuals and give matching score of these two pictures. These scores will be higher when there is a better match. 

content/results.tex

@@ -1,23 +1,46 @@
 \section{results}
 
 \subsection{Overall ROC}
-The overall ROC can be found in Figure \ref{fig:overallROC}. With a false acceptance rate of 0,05, The overall recognition performance is above 99,88\%
+The overall ROC can be found in Figure \ref{fig:overallROC}. To define the false acceptance rate the point on the ROC closest to the northwest corner of the plot. This point is equal to the optimal threshold value. This point results in a false acceptance rate of $0.36\%$ and a correct acceptance rate of $99.62\%$. There is a total of 9358 matches which failed to capture. Together with a total of 383780 matches the failure to capture rate (FCR) is $1.2\%$ assuming that the software only failed to capture on only one of the two pictures. 
 \begin{figure}[ht]
 \centering
-\includegraphics[width=.8\linewidth]{image/final_ROC_log}
+\includegraphics[width=.9\linewidth]{image/overall_ROC}
 \caption{ The overall ROC}
 \label{fig:overallROC}
 \end{figure}
 
 \subsection{Individual ROC’s}
-18 out of the 31 individuals have a perfect ROC. Which means that the highest impostor score is lower than the lowest genuine score. An example of a perfect ROC can be found in Figure \ref{fig:nd1S04934_ROC}. All images from these 18 individuals are analyzed. In all of these images the person looks direct into the camera.  A part of the eyes is visible and no parts of the face are covered. All images have a relative high resolution of 1704~by~2272 pixels. 
+18 out of the 31 individuals have a perfect ROC. Which means that the highest impostor score is lower than the lowest genuine score. An example of a perfect ROC can be found in Figure \ref{fig:nd1S04934_ROC}. All images from these 18 individuals are analyzed. In all of these images the person looks directly into the camera. A part of the eyes is visible and no parts of the face are covered. Not all ROC’s are perfect and further examination will be done later. The results from MATLAB give the feeling that Verilook has a lower success rate for woman. A different ROC in figure~\ref{fig:genderROC} supports this suspicion. There is a clear difference between these ROCs, one should note that the graphs are zoomed. The different scores are split up in table~\ref{tab:gender} and there is a clear difference between the success rate of males and females. This is probably due to the relative small amount of subjects and a Chi-square test gives a p-value of $0.106$. Although the p-value is too high for a significant difference it should be noted that a bigger subject group can generate a significant difference.
+
+\begin{figure}[ht]
+\centering
+\includegraphics[width=.9\linewidth]{image/gender_ROC}
+\caption{General ROC combined with gender specific ROCs}
+\label{fig:genderROC}
+\end{figure}
+\begin{table}[ht]
+\centering
+\begin{tabular}{|c|c|c|}
+\cline{2-3}
+\multicolumn{1}{c|}{} & Male & Female \\
+\hline
+Perfect & 12 & 7 \\
+\hline
+Not Perfect & 4 & 8 \\
+\hline
+\end{tabular}
+\caption{Cross table with gender scores}
+\label{tab:gender}
+\end{table}
+
 \begin{figure}[ht]
 \centering
 \includegraphics[width=.8\linewidth]{image/nd1S04934_ROC}
 \caption{A perfect ROC}
 \label{fig:nd1S04934_ROC}
 \end{figure}
-Not all ROC’s are perfect. In Figure \ref{fig:nd1S04341_ROC} is the ROC of subject nd1S04341. This deviation is caused by one matching score of zero. In figure \ref{fig:nd1S04341} are the corresponding pictures. The main difference between this pictures is the visibility of the ears, further are there no significate difference.
+
+To find out why certain individuals did not get a perfect score it crucial to inspect each of these cases. Figure~\ref{fig:nd1S04341_ROC} is the ROC of subject nd1S04341. This deviation is caused by one matching score of zero. In figure \ref{fig:nd1S04341} are the corresponding pictures. The main difference between this pictures is the visibility of the ears, further are there no significate difference.
 \begin{figure}[ht]
 \centering
 \includegraphics[width=.8\linewidth]{image/nd1S04341_ROC}
@@ -25,12 +48,12 @@ Not all ROC’s are perfect. In Figure \ref{fig:nd1S04341_ROC} is the ROC of sub
 \label{fig:nd1S04341_ROC}
 \end{figure}
 \begin{figure}[ht]
-\begin{minipage}[th]{.45\linewidth}
+\begin{minipage}[th]{.49\linewidth}
 \centering
 \vspace{0pt}
 \includegraphics[width=\linewidth]{image/04341d147}
 \end{minipage}\hfill
-\begin{minipage}[th]{.45\linewidth}
+\begin{minipage}[th]{.49\linewidth}
 \centering
 \vspace{0pt}
 \includegraphics[width=\linewidth]{image/04341d154}
@@ -38,7 +61,8 @@ Not all ROC’s are perfect. In Figure \ref{fig:nd1S04341_ROC} is the ROC of sub
 \caption{Pictures of subject nd1S04341}
 \label{fig:nd1S04341}
 \end{figure}
-Another interesting result is the impact of visibility of the face. A expamle is subject nd1S04406, the ROC of this subject can be found in Figure 6. There are a few genuine scores which a relative low. One of these matching scores in the matching score between the pictures in Figure 6. There are a few differences between these pictures. The first is the visibility of the left part of the face. In the left picture a bigger part of the face is covered with hair. Another significate diffent is the visablity of the ears. Furthermore, only in the left picture, the teeth are shown. 
+
+Another interesting result is the impact of visibility of the face. A expamle is subject nd1S04406, the ROC of this subject can be found in figure~\ref{fig:nd1S04406_ROC}. There are a few genuine scores which a relative low. One of these matching scores in the matching score between the pictures in Figure~\ref{fig:nd1S04406}. There are a few differences between these pictures. The first is the visibility of the left part of the face. In the left picture a bigger part of the face is covered with hair. Another significate diffent is the visablity of the ears. Furthermore, only in the left picture, the teeth are shown. 
 \begin{figure}[ht]
 \centering
 \includegraphics[width=.8\linewidth]{image/nd1S04406_ROC}
@@ -46,16 +70,59 @@ Another interesting result is the impact of visibility of the face. A expamle is
 \label{fig:nd1S04406_ROC}
 \end{figure}
 \begin{figure}[ht]
-\begin{minipage}[th]{.45\linewidth}
+\begin{minipage}[th]{.49\linewidth}
 \centering
 \vspace{0pt}
-\includegraphics[width=\linewidth]{image/04341d147}
+\includegraphics[width=\linewidth]{image/04406d61}
 \end{minipage}\hfill
-\begin{minipage}[th]{.45\linewidth}
+\begin{minipage}[th]{.49\linewidth}
 \centering
 \vspace{0pt}
-\includegraphics[width=\linewidth]{image/04341d154}
+\includegraphics[width=\linewidth]{image/04406d36}
 \end{minipage}\hfill
-\caption{Pictures of subject nd1S04341}
-\label{fig:nd1S04341}
+\caption{Pictures of subject nd1S04406}
+\label{fig:nd1S04406}
 \end{figure}
+
+The pictures in figure~\ref{fig:nd1S04508} have also a relative low matching score. The left picture is a little bit overexposed, making the skin a little bit lighter. Furthermore, the teeth are not visible on the right picture. In figure~\ref{fig:nd1S04508_ROC} is the ROC of this individual. 
+\begin{figure}[ht]
+\centering
+\includegraphics[width=.8\linewidth]{image/nd1S04508_ROC}
+\caption{ROC of subject nd1S04508}
+\label{fig:nd1S04508_ROC}
+\end{figure}
+\begin{figure}[ht]
+\begin{minipage}[th]{.49\linewidth}
+\centering
+\vspace{0pt}
+\includegraphics[width=\linewidth]{image/04508d52}
+\end{minipage}\hfill
+\begin{minipage}[th]{.49\linewidth}
+\centering
+\vspace{0pt}
+\includegraphics[width=\linewidth]{image/04508d71}
+\end{minipage}\hfill
+\caption{Pictures of subject nd1S04508}
+\label{fig:nd1S04508}
+\end{figure}
+
+If the Verilook software was not able to find a face it returned a score of minus one. These negative scores were not used when calculating the ROC but saved in to a different variable. There are eight images in total which have an no-face count larger then 10. Overall do all pictures have a no match rate of 7 or 8 because both pictures get a minus one score if only one of pictures fail to contain a face. Figure~\ref{fig:nd1S04484} shows subject nd1S04484 with a no-face count of 618. While there are 620 pictures in total this picture is matched a total of 619 times which means that only once Verilook was able to recognize a face but failed the other 618 times. In figure~\ref{fig:nd1S04746} a picture of subject nd1S04746 is shown with a no-face count of 395. Again this case Verilook could not detect the face consistently. Analyzing both pictures it is easy to conclude that light a large factor. Lighting is probably the main reason in figure~\ref{fig:nd1S04484} for Verilook to fail even though the subject is looking directly into the camera. Searching for similar pictures where the no-face count is around the 600 it shows settings where the subject wears white clothing. The white clothing results in a lower exposure time which gives a bad lighted face. The subject in figure~\ref{fig:nd1S04746} has a better lighting but does not directly look in to the camera. This leads to presuming that Verilook is depending on random tests to find faces and that this causes a couple of pictures fail only partially.
+
+\begin{figure}[ht]
+\begin{minipage}[th]{.49\linewidth}
+\centering
+\vspace{0pt}
+\includegraphics[width=\linewidth]{image/04484d141}%NoFace:618
+\caption{Picture of subject nd1S04484}
+\label{fig:nd1S04484}
+\end{minipage}\hfill
+\begin{minipage}[th]{.49\linewidth}
+\centering
+\vspace{0pt}
+\includegraphics[width=\linewidth]{image/04746d23} %NoFace:395
+\caption{Picture of subject nd1S04746}
+\label{fig:nd1S04746}
+\end{minipage}\hfill
+\end{figure}
+
+All pictures and all ROC’s can be found in Appendix 1. 

introduction-to-biometrics-paper.tex

@@ -50,7 +50,7 @@
 
 
 
-\documentclass[journal]{IEEEtran}
+\documentclass[12pt,journal]{IEEEtran}
 %
 % If IEEEtran.cls has not been installed into the LaTeX system files,
 % manually specify the path to it like:
@@ -393,8 +393,8 @@
 
 
 % The paper headers
-\markboth{Journal of \LaTeX\ Class Files,~Vol.~14, No.~8, August~2015}%
-{Shell \MakeLowercase{\textit{et al.}}: Bare Demo of IEEEtran.cls for IEEE Journals}
+%\markboth{Journal of \LaTeX\ Class Files,~Vol.~14, No.~8, August~2015}%
+%{Shell \MakeLowercase{\textit{et al.}}: Bare Demo of IEEEtran.cls for IEEE Journals}
 % The only time the second header will appear is for the odd numbered pages
 % after the title page when using the twoside option.
 % 
@@ -431,7 +431,7 @@ The abstract goes here.
 
 % Note that keywords are not normally used for peerreview papers.
 \begin{IEEEkeywords}
-IEEE, IEEEtran, journal, \LaTeX, paper, template.
+FRGC, Verilook, Face recognition, individual ROC
 \end{IEEEkeywords}
 
 
@@ -454,6 +454,7 @@ IEEE, IEEEtran, journal, \LaTeX, paper, template.
 \input{content/method}
 \input{content/results}
 \input{content/conclusion}
+\input{content/discussion}
 
 
 % An example of a floating figure using the graphicx package.
@@ -573,21 +574,17 @@ IEEE, IEEEtran, journal, \LaTeX, paper, template.
 %
 
 
-\appendices
-\section{Proof of the First Zonklar Equation}
-Appendix one text goes here.
+%\appendices
+%\section{Proof of the First Zonklar Equation}
+%Appendix one text goes here.
 
 % you can choose not to have a title for an appendix
 % if you want by leaving the argument blank
-\section{}
-Appendix two text goes here.
+%\section{}
+%Appendix two text goes here.
 
 
 % use section* for acknowledgment
-\section*{Acknowledgment}
-
-
-The authors would like to thank...
 
 
 % Can use something like this to put references on a page
@@ -620,13 +617,13 @@ The authors would like to thank...
 % <OR> manually copy in the resultant .bbl file
 % set second argument of \begin to the number of references
 % (used to reserve space for the reference number labels box)
-\begin{thebibliography}{1}
+%\begin{thebibliography}{1}
 
-\bibitem{IEEEhowto:kopka}
-H.~Kopka and P.~W. Daly, \emph{A Guide to \LaTeX}, 3rd~ed.\hskip 1em plus
-  0.5em minus 0.4em\relax Harlow, England: Addison-Wesley, 1999.
+%\bibitem{IEEEhowto:kopka}
+%H.~Kopka and P.~W. Daly, \emph{A Guide to \LaTeX}, 3rd~ed.\hskip 1em plus
+%  0.5em minus 0.4em\relax Harlow, England: Addison-Wesley, 1999.
 
-\end{thebibliography}
+%\end{thebibliography}
 
 % biography section
 %