• file
• revisions
• annotate
• diff
• raw

reports/final/content/finalproduct.tex@7f671b74a3a8 (annotated)

reports/final/content/finalproduct.tex

2010-03-19

author

francis

date

Fri Mar 19 09:09:52 2010 +0000

changeset 395

7f671b74a3a8

parent 393

352b1179b4f7

child 396

f190851b2dbb

permissions

-rw-r--r--

Other stuff

Willy@320	1	\chapter{Conclusions}
Willy@320	2	\section{Final Product}
Dragos@371	3	The final prototype is a functional iris detection application, which provides the
Tom@375	4	user with visual prompts and feedback and also some supplementary information about
Tom@375	5	the loaded image for debugging purposes.
francis@318	6
Tom@375	7	The final product can read in an 8-bit indexed, greyscale image and will attempt to
Tom@375	8	automatically best-fit parameters for the location of the pupil, iris and eyelids with
francis@395	9	consideration for specular highlights and padding for eyelash interference. The entire auto-detection process takes a matter of milliseconds. The user may
francis@395	10	choose to manually enter points to specify all of these locations; if the user does, he
Tom@375	11	or she will receive appropriate visual feedback cues for repositioning elements in a
Tom@375	12	manner intuitive to any casual computer user.
francis@318	13
francis@395	14	From this point, the user input is complete and the application will process the
francis@395	15	image data, unwrap the iris using polar co-ordinates and mask appropriate
francis@395	16	regions of the unwrapped iris based on the auto-detection sequence or user
francis@395	17	input. This information is then extracted using Gabor wavelet convolution
francis@395	18	filters and the final output is a 2048-bit iris code containing phase
francis@395	19	information of the valid iris pixel data.
Dragos@371	20
francis@395	21	The iris code is stored in a local, new line delimited text database if unique
francis@395	22	or the prototype reports that the code has already been located in the database
francis@395	23	and a successful match dialog is given.
francis@318	24
francis@395	25	\section{Auto Detection}
francis@318	26
Tom@375	27	%\subsubsection{Pupil}
francis@318	28
francis@395	29	Pupil detection in the application works to a very high degree of accuracy in
francis@395	30	every seen case. The pupil has a fairly unique shade in comparison to the rest
francis@395	31	of the eye and its surrounding area; this enables an intelligent threshold to
francis@395	32	be carried out with information from the image histogram to isolate the pupil.
francis@395	33	This, unfortunately, is not a property shared by the iris, making it significantly more
francis@395	34	difficult to isolate than the pupil.
francis@318	35
Tom@375	36	%\subsubsection{Iris}
Willy@331	37
Tom@375	38	Currently the prototype makes an estimation of the location of the iris based on
francis@395	39	concentricity (a potential source of error) with the pupil, and assumes
francis@354	40	a detectable gradient ramp between the iris and the sclera. These complications
francis@354	41	hamper the accuracy of our current estimates and produce a successful detection
Tom@375	42	rate on the CASIA database of around 70\%, though potential improvements to this
Tom@375	43	method are suggested in future work (\ref{furtherwork}).
Willy@343	44
Tom@375	45	%\subsubsection{Eyelids}
francis@360	46
Tom@375	47	The prototype manages to detect the top edge of the eyelids in the vast
francis@360	48	majority of cases. With the small amount of padding added to accommodate the
Tom@375	49	thickness of the top eyelid, it can be seen that, although not perfect, a very
Tom@375	50	reasonable estimate of the eyelid boundary is found automatically.
francis@360	51
francis@393	52	\chapter{Final Results}
Willy@343	53
francis@393	54	With the batch command line option of the application, it is possible to
francis@393	55	process a large amount of images of eyes. Using this option, 108 images of eyes
francis@395	56	from the CASIA database were loaded into the application; the pupil, iris and eyelids were auto-detected to generate a bitcode and store it in the database. These were then compared to an alternative image of each eye
francis@393	57	to check for matches.
francis@393	58
francis@393	59
francis@395	60	To further test primarily for false matches, the first three images of each eye were
francis@393	61	loaded into the database, and then compared to a fourth
francis@393	62	image for each eye. As there were originally three images for each eye, and
francis@393	63	then a comparison was performed on all of the extra 108 images, roughly 35,000
francis@393	64	iris comparisons took place ($(108 * 3) * 108$).
francis@393	65
francis@393	66	Our results of these tests are as follows: \begin{itemize} \item 0 false
francis@395	67	matches in $\sim$35,000 comparisons \item 70\% (75 out of 108) match rate
francis@395	68	with Professor Daugman's suggested 0.32 Hamming distance.
francis@354	69	\end{itemize}
Willy@343	70
francis@360	71	The complete lack of any false matches and the incredibly high match rate are a
francis@393	72	testament to the robustness and feasibility of an iris recognition system. The
francis@360	73	tests have suggested that in virtually any case where the iris boundary is
francis@393	74	correctly located, the iris should be correctly identified.
francis@356	75
francis@393	76	As a roughly 70\% match rate was achieved for the iris location, this percentage has, as
francis@393	77	predicted, been reflected in the overall match rate in the
francis@395	78	database of images. Similarly to Daugman, we can also observe that the Hamming
francis@393	79	distances produced by comparing different irides tends to follow a binomial distribution, with a mean around 0.46; see
francis@360	80	figure \ref{hd}.
francis@354	81
Willy@363	82	\begin{figure}
francis@356	83	\centering
francis@356	84	\includegraphics[width=0.55\textwidth]{hd}
francis@356	85	\caption{The distribution of Hamming distances for one run of all the irides in the database}
francis@356	86	\label{hd}
francis@356	87	\end{figure}
Willy@343	88
francis@393	89	\newpage
francis@393	90
francis@359	91	\section{Source Code}
francis@393	92	The source code is freely available under the GPL licence from \url{http://projectiris.co.uk}.
francis@359	93
francis@393	94	\chapter{Future Work}
Tom@373	95	\label{furtherwork}
Tom@375	96	Throughout development the focus was on producing a functioning piece of software capable of
Tom@375	97	fast, consistent and reliable iris recognition. There are other possible routes for extension
Tom@375	98	and exploration within this project. There are alternative ideas for the format of the
Tom@375	99	application and unexplored avenues within auto-detection techniques.
Willy@343	100
francis@393	101	\section{Otsu Thresholding for Iris/Sclera Separation}
francis@359	102	As we have seen before on average the iris has a lower intensity than the
francis@359	103	sclera. Hence it could potentially be beneficial to explore the use of Otsu
francis@359	104	thresholding for separation of the iris and sclera. Similar to the methods
francis@359	105	implemented for eyelid detection, if we remove eyelids and pupil intensities
francis@359	106	from the Otsu thresholding process then we obtain the image shown in figure
francis@359	107	\ref{otsu-iris}. Although to the human eye this image looks to hold more
Willy@363	108	information than the implemented technique (see figure \ref{iris-threshold})
francis@359	109	the image contains a large amount of noise and would require further filtering
francis@359	110	and refinement to be useful.
Willy@343	111
Willy@363	112	\begin{figure}
Willy@343	113	\centering
Willy@343	114	\includegraphics[width=0.35\textwidth]{anothereye}
Willy@343	115	\hspace{1cm}
Willy@343	116	\includegraphics[width=0.35\textwidth]{otsui}
Willy@343	117	\caption{Image before and after an Otsu threshold.}
Willy@343	118	\label{otsu-iris}
Willy@343	119	\end{figure}
Willy@343	120
francis@359	121	This would be the first recommended route for extension since, during the
francis@359	122	testing phase, the software tended to fail recognition on images where it could
francis@359	123	not adequately locate the iris.
Willy@344	124
francis@395	125	\section{Eyelashes}
francis@359	126	In some images eyelashes can obscure parts of the iris. Detection and removal
francis@359	127	(marking as bad bits) of eyelashes could potentially improve the detection
francis@359	128	rate. A method is given in \cite{min2009} for eyelash detection. The method
francis@359	129	relies on two characteristics of eyelashes: firstly, that they have a relatively
francis@359	130	low intensity compared to surrounding features, and secondly that the eyelash
francis@359	131	regions have a large variation in intensity due to the thin shape of the
francis@359	132	eyelashes (see figure \ref{justaneye}). The method uses these characteristics
francis@359	133	by combining a standard-deviation filter and a thresholding technique.
Tom@329	134
Willy@363	135	\begin{figure}
Tom@329	136	\centering
Tom@329	137	\includegraphics[width=0.35\textwidth]{justaneye}
Tom@329	138	\caption{The eyelashes have a relatively low intensity and the eyelash region has high variation in intensity.}
Tom@329	139	\label{justaneye}
Tom@329	140	\end{figure}
Tom@329	141
francis@395	142	\section{Eyelids}
francis@359	143	The implementation of eyelid detection is based on the region obtained from the
francis@359	144	thresholding technique, as we saw in figure \ref{otsu-eyelids}. The system
francis@359	145	currently constructs a parabola from three points on the boundary of this
francis@359	146	region. Accuracy in eyelid detection could potentially be improved by using the
francis@359	147	Hough technique for parabolas. This again would allow more accurate marking of
francis@359	148	non-iris regions and hence improve overall recognition rate.
francis@318	149
Willy@344	150	%TODO:? \subsection{Gaussian Derivative Encodings}
francis@318	151
francis@395	152	\section{Application Format}
francis@359	153	Currently the system has a text file for storing iris encodings. This could be
francis@359	154	improved to mirror a real-world application deployment by interfacing with a
francis@359	155	database. Although due to the one-to-many searches involved with iris
francis@359	156	recognition database querying could potentially become a bottle neck.
francis@318	157
francis@359	158	Another area worth exploring is implementing the software with a client/server
francis@359	159	architecture whereby the iris is encoded client side, and compared server side.
francis@359	160	Again, this more accurately mirrors real-world deployment of an iris recognition
Willy@363	161	software. This would then require further investigation into security and encryption methods
Willy@363	162	for the system.
Willy@344	163