0:00:16so my name is patrick bob um from france of a little uh uh and also france
0:00:22and so this talk is gonna be a
0:00:25uh what i'm knocking security a mostly out to design secure well uh and baiting scheme
0:00:31and then will show you how to use a nice mess medic zero record optimal to once per in order
0:00:36to minimize the distortion and and also to guarantee what um
0:00:44so the line is very simple a first introduction on the
0:00:48and what are mocking security is and uh i will uh
0:00:52propose requirement in order to achieve secure and baiting
0:00:56uh i would propose so different embedding scheme and compare them using a performance and that is this
0:01:02and then will conclude my work
0:01:04after the what
0:01:06okay uh what's mean security in what are marking so a security as thing to do is do you picked
0:01:13compression uh no what at to share additional like
0:01:16go and uh
0:01:20uh you have to consider a first and that this ad that's every
0:01:23it's is very important because the are just every as the brain
0:01:26is able to
0:01:28try to hack you'll system to perform at that
0:01:31it can i have a a a really do a an immediate problem in a uh a a where E
0:01:36it is gonna depend of your uh a now you that you're gonna can see there
0:01:40but uh this is some ones that would try so two
0:01:45hard your system
0:01:47that that's sorry you have to consider that yeah as a bunch of my tell you or
0:01:52than that that you can be for example the what tell the code they'll detect detector then you would be
0:01:56able to yeah form more what we code or equal attack
0:02:01here we go that can see there's that's the
0:02:03at necessary will
0:02:05have i con turns that that what they mapped to read the
0:02:08and a a what a map with the same key
0:02:11you can have all of L scenario where you assume that the just every for example knows the ms that
0:02:17are embedded in so but a map content T pretty depends of the signal you
0:02:21and a strong assumption is that uh you assume that to what am is a um uh
0:02:26that that sorry know everything about
0:02:30but they marking scheme except the secret key
0:02:33this this is also one the or so as assumption
0:02:36and what are than the is objective of this uh the adversary re
0:02:40in all cases than a be to estimate you secret key and uh if it's it's to make what they're
0:02:47are all copies of sage
0:02:49and that got that don't actually a lot about what is going to a a a a what the is
0:02:53that the stories is uh doing to do because it in all case we can that
0:02:57the uh uh design secure embedding scheme
0:03:01so uh
0:03:03here are the assumption that we are are using
0:03:06we can see they are that the almost is good shown
0:03:09uh is C is this makes sense to sends to the some limit though is your M of course
0:03:15um and there's all assumption is the fact that uh as a horse is i I D in the
0:03:20in there
0:03:22i and in
0:03:23plea uh distributed
0:03:25we assume that all the content uh what them up using the same secret key
0:03:30and is this means that there is a and that that board W A a that map content only at
0:03:36that so we have in this frame
0:03:39the um
0:03:39i just necessary
0:03:41uh as only what am i can turn to does not know about them but didn't nice H for example
0:03:49or was the or here out to design uh secure scheme
0:03:53uh and also sorts all to play with a to was also a very strong can trends and what that
0:03:58masking meanings the
0:04:00and weighting distortions that has to be minimised
0:04:03and also to maximise the robustness
0:04:09i to minimize the or uh let's and the robustness so yeah we use um
0:04:15the rationale already use since a long time in what they're marking
0:04:19it's to use in it coding and you set side information from the almost
0:04:23is come from the famous close that's paper but one dove T paper or uh uh
0:04:28a what that marking
0:04:29and uh the I Ds and an I D to generate a different coding rage and that we'll code for
0:04:35the same may say
0:04:38is an baiting in very simple in this case since you have different and dating regions in a to embed
0:04:43them a said you go to the close
0:04:45decoding regions and you
0:04:47uh uh you're able to minimize the distortion okay
0:04:51so no i will a some up the stream main we got crime i'm and fall secure them baiting
0:04:57so the first one that i called it distribution splitting
0:05:02you are have uh
0:05:03you want to achieve a
0:05:05a very strong
0:05:07statistical property
0:05:09chord is a perfect secrecy uh in steak and the refugee of take was security in what they marking
0:05:15it's means that the distribution of the content
0:05:18uh is the original content so P of X and the distribution of the what um not content i it
0:05:24E E Y given the key a exactly the same
0:05:28if you are of these properties and so just a re can not do anything about a uh i can
0:05:34that that that you'll system then because it doesn't know is a coefficient are what all marked or even
0:05:39uh is the carrier may age
0:05:44here we assume that we are doing a binary and baiting in means that actually you can't really
0:05:50you're your uh distribution of what that map the content
0:05:54P of Y into two distribution
0:05:57the one uh for of the ms a a zero and the one for the said one
0:06:02and of course you have to divide
0:06:04a two in order to normalize the distribution
0:06:07uh in a bill to achieve this predicting we go now use a partitioning function uh J G
0:06:15and uh you will see i would show you an example but basically once you have G A a a
0:06:20uh G
0:06:21you can compute the
0:06:24distribution and fall zero according to this partitioning function then you can compute the D
0:06:30be shown the what are not good content and bidding one
0:06:33also according
0:06:34to this partitioning function
0:06:37you're is an example of the J function so it's
0:06:43that a wise uh function
0:06:46and so you see that the and function your is divided so it
0:06:50in in two parts
0:06:52here i just
0:06:54are you
0:06:55the part for one dating zero but you have also as you a part one bit one so the blue
0:07:00K here
0:07:01is the distribution
0:07:02one you want but you want but you
0:07:05if you add the two distributions of the chrome complain military terry one for dating one you of the now
0:07:13exactly the send distributions and the distribution of the also
0:07:19so it is is the first requirement distributions splitting
0:07:22now we have to find a map being in of the two
0:07:26go from
0:07:27the distribution of
0:07:28well can which is gauche shown
0:07:31with a assumption
0:07:32to is the new distribution when you want for example to embed bet one okay
0:07:38we can find plenty of different mappings of course
0:07:43we can that holds this mapping being uh function T get beat L T
0:07:47and so the requirement
0:07:49as a already say so
0:07:51you have to be to to this
0:07:54so the requirement is that you want to find this because the mapping that will minimize embedding dating distortion
0:08:01would do we so
0:08:02so this is important because you have
0:08:04but you can invent of different mapping but is only one that's
0:08:09will minimize the average at two T distribution
0:08:13and here we use optimal once spot in order to do this
0:08:17so the optimal tons pulse you re give a uh implicit form we'd out of the being in the is
0:08:23the case of a scan a distribution so in in one D
0:08:27it's a given by the
0:08:29mean it
0:08:30relative distribution function cat be tied at all of the density of
0:08:34the what that mark content
0:08:38that is applied on the cumulative distribution function of the host content get beat that F four S
0:08:45this is the that
0:08:46the mapping from your
0:08:48you can also derive a is the and bidding being distortion it's right then here
0:08:54and so you can
0:08:55once you know the distribution and you can compute the mapping for
0:08:59he distribution uh you you shows them
0:09:03uh we're is an example so still a my oldest below
0:09:07the be shown than the target distribution and one dating is you were
0:09:11you a a one example of such a mapping here
0:09:15so this mapping being will and able to minimize at this
0:09:18the the distortion on average
0:09:23is that everything is done you you you you you you can choose different partitioning function of your distribution you
0:09:30know either
0:09:31to uh that form baiting
0:09:33this is what was called the a once upon that you are what our marking
0:09:38so you the distribution is very simple if you want to one bad you will all the cool quite if
0:09:43what are not
0:09:44a coefficient of an i on the left side of the potion distribution
0:09:48if you want to embed one
0:09:50it gonna be on the right side
0:09:52this is not in at because you have only one a coding region of dictionary region
0:09:58uh and it was a it comes from a use what
0:10:02uh no or you can
0:10:04play with is the distribution and design new distribution here is an example where a or coding and as the
0:10:11set an uh are and we also the same probability
0:10:16so the the a here is the same than this one and so long
0:10:19so i called E
0:10:21he not you are what that marking be close all
0:10:25the distance is the
0:10:27region at the same probability P
0:10:29in all the two
0:10:32and as its and probably T P yes
0:10:35uh you can also see me try a seem that tries this uh D
0:10:39the and then you are
0:10:41P bar
0:10:42that you are what they marking it exactly the same than the previous one but
0:10:46i perform a tree according to a a or or uh around zero
0:10:52yeah again i managed to a of the pleading distribution requirement a and and they able to to from and
0:11:01the last
0:11:02partitioning function i tried that
0:11:05is called
0:11:06the no
0:11:07that natural are what a marking where a or or then beating uh region as the same way
0:11:14that is code then does so it's very similar you know to this
0:11:18can of post test scheme all to you a i am but on you know you have this uh secure
0:11:24and bidding in this case
0:11:28so i will now uh compels of defer and to uh what a mapping method the a regarding uh a
0:11:35tell of eight okay
0:11:37i want to evaluate knows the last constrains the robustness so i will uh i'm but now assume uh additive
0:11:44white caution noise uh channel
0:11:47and here is the compare is and between the that than that you are what that marking and france pop
0:11:52net you are what that marking needs
0:11:54to evaluate
0:11:55the benefit of doing
0:11:57inform it coding
0:11:59and as you can see so
0:12:00he's low well well a plot is for
0:12:04that than that you are what that marking so you can see that you can that shape
0:12:08uh small of bit error rate
0:12:10once you can see that in front of coding spatially for lower that uh the value you in a
0:12:16the you and a the what them map and of the ratio of cool
0:12:22this is so of first benefit
0:12:24no if i can they so as a different
0:12:27uh and they so that than that you to what they marking P that you are what that mapping and
0:12:32P bound that you are what yeah marking
0:12:35again a again but if they are and what are marked to knows the ratio
0:12:38you can see is that for example in this case we have a what they map to content which sure
0:12:43of minus five db
0:12:45in this case so first the P uh and that you are without marking
0:12:49gives the best yeah formants below minus five db and that the are you have to use the P bound
0:12:56that you are what that marking in of the two
0:12:58uh degrees
0:12:59you're a bit error rate
0:13:02i i degrees again to what that mark to content issue
0:13:06i i uh no sorry yeah
0:13:09i wanted also to compare now we is and secure but known to be robust what the marketing scheme so
0:13:14you have to mail for
0:13:16to more plots
0:13:17is that
0:13:18got more less picky read for uh in bits picked one here
0:13:22so you can see for example that the improve
0:13:24a like all for of course low well bit error right for the value and and
0:13:29and this get us but that's scheme E which fill also
0:13:33low low well it or right so
0:13:36is this is a kind of a known in in the secure what they marking it's very out to be
0:13:42both robust and very secure
0:13:44this the rubber scheme yeah wow
0:13:47a more robust and the secure one
0:13:50and knows of embedding distortion and minus
0:13:53eleven D V yeah you can see is that
0:13:56main is the
0:13:58uh that than that you are what that marking out their forms
0:14:01the was a a a a uh and dating scheme
0:14:05so the bit the it much real well for this one
0:14:08if again i compare our we is this can ask what S T in uh and uh
0:14:13is the input speech spectrum
0:14:15i can see is that
0:14:17the the kind of "'cause" that's in give
0:14:20even beta a bit error rate
0:14:22even if
0:14:24as a at the the value and they to zero you are very you are very close performance as
0:14:32it is also interesting
0:14:34no i will conclude group michael so
0:14:40it's its importance so to perform distributions coefficients so to split your this
0:14:45or and you know up to achieve a
0:14:46secure and baiting
0:14:48you have also to find a way to match the two distributions so the one of the all sins one
0:14:52of the what they not go though
0:14:55in order to do so uh a to do this you can use of to much from sports
0:14:59it's very it's walks fine in one dimension mention
0:15:02is the problem is that in a a multi dimensional it's more complicated but you can use all also of
0:15:08optimization trees and mean know up to minimize the distortion
0:15:12uh there is a gain of using in from it in also for secure and baiting
0:15:17and uh and thought and that is so the best partitioning depends of the and dating
0:15:23used option the but use yeah and also of the noise
0:15:26so i
0:15:27we don't have a a a a a magic old up to perform both secure and and and dating
0:15:34the best pick either a fell to link
0:15:37what we can do using secure adaptation of the get us but that's you
0:15:41so they have been applied by around the last week and is using at at at that position of a
0:15:46uh this yeah that's but that's scheme
0:15:49and they are or so a more fundamental problems as like i would to compute exactly theoretically is a secure
0:15:55get by is so
0:15:56how much information they can convey a while keeping being uh security
0:16:02and also
0:16:04another problem is that
0:16:06if you want to what secure embedding betty a secure uh and baiting uh
0:16:10with is low bit error rate you have also to use secure uh iraq weighting
0:16:15yeah well only can code
0:16:17and this is uh not known right now also we have to find a way to
0:16:23improve the quality of cell is but also guarantee security it's not easy because if you use uh
0:16:29they're all calling
0:16:30encode you gonna add dependence between the
0:16:33symbols that you're gone and that
0:16:35and this dependence can be a security uh equal
0:16:39thank you for your attention
0:16:45and the time for a couple
0:16:52that to a kind of nice the
0:16:54didn't on the first like to set you word and work on a C is just a done the quite
0:16:58sure what type of security does your
0:17:01net the dress
0:17:03and and is you cure in the first place because you kind of
0:17:06i use you could in first two slice
0:17:08you never argue why he's
0:17:10Q also okay
0:17:11so uh
0:17:13here are the signal you is that so that very will have only what a market the content and we
0:17:19try to
0:17:21it's steam so you will have a bunch a
0:17:23what a market contents
0:17:25all more more to be uh what L mark only a set of then will be what they matt
0:17:30and we assume that all this can this as a cushion and and i a E okay
0:17:35so the goal of that the we will be two
0:17:38for example is to the set
0:17:42that uh
0:17:45so what am a
0:17:47so if you want to do this
0:17:50i i has to use some kind of uh uh security at that it can be for example looking at
0:17:56the distribution
0:17:57fine cluster
0:17:58yeah from independent can it uh and component and that is this would you have
0:18:02very very different way to perform from uh what uh what am looking at that in the security uh a
0:18:08lot at the security level
0:18:11we uh i assume that
0:18:14is there is not possible at that
0:18:17all the coefficient have the same distributions so what it is that the sre
0:18:21won't be a bore
0:18:22to know which coefficient
0:18:24i what alma
0:18:27and still not quite a what what
0:18:29what is the goal of the attacker
0:18:31i'm still not getting a
0:18:33so what is that the vet analysis that you trying to
0:18:35event of time
0:18:36so it's to if you at this you know you a few out of the N coefficient and only you
0:18:42set of and coefficient
0:18:44conveys or information
0:18:46is the N coefficients the location of this and clean fish and uh come from us
0:18:51secret key
0:18:52so the of the idea so you will be two
0:18:55uh try to estimate
0:18:58which coefficient fusion
0:18:59conveys or information
0:19:03and then then the next
0:19:05why that's the case why is this in system because it didn't see any secure you know
0:19:09i only solar boston's in
0:19:12yeah the security is grounded because you assume
0:19:15because uh with the em baiting
0:19:18you have exactly the same distribution before and after of baiting
0:19:21it's means that
0:19:22if i want if i plays a but yeah i and and the i am that that's every
0:19:26i won't be a ball
0:19:28to locate
0:19:29which coefficient carries a what that matt so if i want to
0:19:33to try if i tells a say J you have to add the noise on all the coefficient of my
0:19:40i cannot say act
0:19:41which one carries a a age
0:19:45as a a just every
0:19:50a left so offline
0:19:56so thank you know match again