ANIMATION OF 3D MODEL OF HUMAN HEAD

58 J. MIHALÍK, V. MICHALČI, AIMAIO OF 3D MODEL OF HUMA HEAD AIMAIO OF 3D MODEL OF HUMA HEAD Ján MIHALÍK, Vikto MICHALČI Laboatoy of Digital Image Pocessing and Videocommunications, Dep. of Electonics and Multimedia elecommunications, FEI U Košice, Pak Komenského 3, 04 0 Košice, Slovak epublic Jan.Mihalik@tuke.sk, michalcin@vk.upjs.sk Abstact. he pape deals ith the ne algoithm of animation of 3D model of the human head in combination ith its global motion. he designed algoithm is vey fast and ith lo calculation equiements, because does not need the synthesis of the input videosequence fo estimation of the animation paametes as ell as the paametes of global motion. he used 3D model Candide geneates diffeent expessions using its animation units hich ae contolled by the animation paametes. hese ones ae estimated on the basis of optical flo ithout the need of extacting of the featue points in the fames of the input videosequence because they ae given by the selected vetices of the animation units of the calibated 3D model Candide. he established multiple iteations inside the designed animation algoithm of 3D model of the human head beteen to successive fames significantly impoved its accuacy above all fo the lage motion. Keyods Animation, 3D model, human head, animation paametes, global motion, local motion, optical flo, estimation. Intoduction MPEG-4 SHC [] uses the ne methods of coding of the human head based on its modeling [] and animation [3]. he animation enables the local motion of 3D model of the human head by using animation paametes including its global motion. he animation paametes can be geneated synthetically o they ae the esult of analysis of the eal human head. Except fo the global motion as otation and tanslation the human head in the input videosequence also has the local motion hich is caused by a face mimic. he typical case of the local motion of the human head is opening of its mouth at geneation of the speech. hen animation in the model based human head coding [4] geneates by the local motion of 3D model the diffeent expessions of the synthesized human head fo vebal and nonvebal communication. he method of coding is an impotant pat of the standad MPEG-4 SHC, hich allos such applications as the advanced communications using the cloned o vitual human heads. he local motion o mimic of 3D model of the human head can be classically expessed by the method of clip-and-paste [5] fo hich the stuctue of 3D model is not defomed. Fist the textue of chaacteistic pat like mouth and eyes of the human face is cut out fom the fames of the input videosequence. he local changes of the human head, hich paticipate on animation have to be expessed in egad to the textue. We can see the textue as an image of the pat of the human face hat means that the vecto quantize [6] can be used fo its coding. he code book of the vecto quantize involves the templates of the pat ith diffeent expessions. hen by using the code book e can geneate the finite numbe of possible expessions of the pat of the synthesized human head. Moe complex expessions not fo a pat but fo the hole synthesized human head ae geneated by combination of seveal templates fom the code books of the diffeent pats hich have to be tanslated to the decode. heefoe the code books include the big numbe of templates the method of clip-and-paste is not vey good fo lo bit ate coding of the human head. In addition the expessions have to be pasted on the ight places in the synthesized fames hat is not simple solution. Even small inaccuacy of thei pasting affects busy in the synthesized fames. Bette solution fom the point of vie is mimic expession by animation of 3D model of the human head on the basis of featue extaction and tacking [7]. he impotant featues like points, edges and contous of the eal human head in the input videosequence can be extacted by the defomable templates [8] of its pats o by application of the pincipal component analysis [9] on the ones. he method of optical flo [0], [] by using the paameteized 3D model of the human head enables excluding of the need of extaction of the featues fo its animation... Paameteized 3D model Candide Each vetex M = (h,v,) of 3D model Candide [] has the initial position in the model coodinate system (MCS) defined by the hoizontal (h), vetical (v) and depth () codinates. he coodinates of its all vetices ae nomalized on the maximum value. he advantages of

RADIOEGIEERIG, VOL. 6, O., APRIL 007 59 3D model Candide ae its availability and mainly simplicity efeed to the elative small numbe of vetices and polygons. ext its advantage is that almost all featue points of the human head defined in the standad MPEG-4 SHC have in 3D model Candide the coesponding vetices. 3D model Candide is paametized by the folloing equation x= x+ us+ A () hee x is 3 dimensional vecto composed fom 3D coodinates of its vetices x 3D model in the initial position S matix of shape units A matix of animation units u shaping paametes animation paametes he numbe of columns of the matices S and A depends on the numbe of shaping and animation unites, espectively. he diffeence beteen shaping and animation units is that the shaping units define defomation of the geometical shape of 3D model Candide hile the animation units define its local motion in dependence on a mimic of the eal human head. Fo the global motion of 3D model Candide given by otation and tanslation of its all vetices also its scaling thee ae needed next 7 paametes. hen eq. () changes as follos ( ) x= mr x+ us+ A + hee m is a scale, R = R(Θh, Θv, Θ) is the otation matix ith Eule s otation angles Θh, Θv a Θ and = (th, tv, t ) is 3D tanslation vecto Fom eq. () follos out that shaping, animation and global motion including scaling of 3D model Candide ae descibed by the paametic vecto [,u,] [ h v h v ] () D =ν = Θ,Θ,Θ,t,t,t,m,u, (3) hee ν is a vecto of the global motion paametes and scale. A change of the geometical shape of 3D model Candide can be caied out by using the shaping units (SU) hich numbe is 4. Each SU is labelled by its name deived fom the kind of geometical shaping. Animation of 3D model Candide allos intepetation of diffeent local motion. In geneal the local motion can be ealized by the animation units (AU) hich numbe is. Each AU is identified by its name deived fom the kind of local motion and is defined by a gope of vetices, hich enable the selected animation. he vetices of AU have animation vectos, hich expess the maximum changes h, v, of thei coodinates fom the initial positions in MCS. In Fig.a,c fo 3D model Candide thee ae outlined the vetices togethe ith thei animation vectos fo AU of Raising uppe lip and Dopping of ja. he maximum changes as components of the animation vectos ae shon in ab. and fo both AU, espectively. ab.. AU Raising uppe lip Vetex Animation vectos indices h v 7 0,000000 0,086957 0,0739 33 0,000000 0,0657 0,0739 66 0,000000 0,0657 0,0739 79 0,000000 0,050000 0,0739 80 0,000000 0,050000 0,0739 8 0,000000 0,050000 0,0739 8 0,000000 0,050000 0,0739 87 0,000000 0,0657 0,0739 ab.. AU Dopping of ja Vetex Animation vectos indices h v 40 0,000000-0,60000-0,050000 8 0,000000-0,60000-0,050000 9 0,000000-0,60000-0,00000 0 0,000000-0,30000-0,50000 3 0,000000-0,50000-0,30000 65 0,000000-0,50000-0,30000 83 0,000000-0,00000-0,050000 84 0,000000-0,00000-0,050000 85 0,000000-0,00000-0,050000 86 0,000000-0,0000-0,050000 hen fo the coodinates of the animated vetex M ' = (h',v',') in MCS ae valid h' h h v' = v + v ' hee is the animation paamete, hich by its value fom the inteval 0, detemines the quantity of thei changes fom the initial values (h,v,). It means that each AU has one animation paamete by using hich its all vetices ae cuently affected. In Fig.b thee is shon the animated 3D model Candide by using AU of Lifting uppe lip and in Fig.d by AU of Dopping of ja assumed the maximum value = fo both AU. Fo animation of 3D model Candide it is possible to use seveal AU hich animate its diffeent pats. he esult such a combination of to befoe given AU is an integated expession as it is shon in Fig.e. In addition, to vetices of one AU can belong to the diffeent animation vectos fom hich follo out that one AU can intepet moe expessions of the human head. he fact is poved by the simple example hen AU fo shutting of the lid ill be the same as AU fo its full closing but thei animation vectos ae diffeent. (4)

60 J. MIHALÍK, V. MICHALČI, AIMAIO OF 3D MODEL OF HUMA HEAD a) b) e) c) d) Fig.. Animation of 3D model Candide by using AU Raising uppe lip and Dopping of ja a), c) the animation vectos, b), d) the animated 3D model by the sepaated AU e) the animated 3D model by combination of both AU. h x' x + x 3. Animation ith small local motion = (8) z' z he local motion of 3D model Candide is defined by z AU, hich ae affected by the animation paametes W = (,..., n ). ext e deive a fomula fo v + estimation of the animation paamete fo any AU y y' y = (9) assumed the small local motion beteen to successive z' z fames. he fist fame of the ones is a efeence fame z unde hich 3D model Candide and the camea ae calibated and the local motion is descibed by eq.(4). All ext afte multiplying and by using the expession of vetices of AU fom the point of vie of estimation of pespective pojection [3] fo the components (u i,u j) of can be consideed as the featue ones and thei pojections the vecto of optical flo e have on the fame as the featue points. hen the coodinates of the animated featue vetex afte thei tansfomation to the J + fx h camea coodinate system (CCS) [3 ] ill be u j = z (0) h x' = x + x (5) z I fy v v y' = y + y (6) u = z i () z z' = z z (7) hee I = ( i i 0),J = (j j 0) ae the cented coodinates of the featue point in the fame. Eq.(0) and () epesent a If eq.(5) and (6) ae divided by eq.(7) e get non linea dependence of the components of the vecto of optical flo on the animation paamete. Assuming small local motion beteen to successive fames ( <<) and the big distance d of the camea fom 3D

RADIOEGIEERIG, VOL. 6, O., APRIL 007 6 model Candide compaed to its depth coodinates fo the featue vetex can be done by next equation denominato of eq. (0) and () is valid ˆ y x I I f v ˆ i I J + f h + j () = Î t (8) z hee the patial deivatives ˆIi = I(i+ u ˆi,j+ u ˆ j,t+ )/ i, hee z=d-. Afteads fom eq.(0) and () in egad to eq. () the linea dependences of the ˆIj= I(i+ u ˆi,j+ u ˆ j,t+ )/ j and the time diffeence components ( u i,u j) of the vecto of optical flo on the ˆI ˆ ˆ t = I(i+ u ˆi,j+ u ˆ j,t+ ) I(i,j,t) Iiuˆi Ijuˆ j. In the animation paamete ae obtained expessions the pedicted components of the vecto of J + fx h optical flo ae uj = (3) J + fx h ûj = ŵ (9) I fy v ui = (4) I fy v ûi = ŵ (0) By substitution of eq.(3) and (4) into the optical flo equation [4] e get fo the featue vetex one linea hee ŵ is the animation paamete fom the pevious equation fame as it is seen in Fig.. I fy v J fx h I i + I + j = I t (5) Coection W hen fom eq.(5) the animation paamete fo the small of animation W local motion can be diectly calculated. Fo bette paametes estimation of the animation paamete it is needed to use moe featue points above all fo AU ith bigge numbe Pediction of the vetices of 3D model Candide. In matix fom of animation eq.(5) fo featue vetices can be itten as follos paametes p ep It M = M (6) p e p It y x hee e I f v J + f h = I i + Ij and I ae t membes fo the featue vetex,,,... On the basis of the least squae method (LSM) a solution of eq.(6) fo the animation paamete of the small local motion ill be = e e I t (7) In eq.(6) the maximum numbe of the selected featue vetices is equal to the numbe of the vetices of AU. o achieve highe accuacy of estimation of the animation paamete fo the given AU it is suitable to take all its vetices. 4. Animation ith lage local motion In geneality fo estimation of the animation paametes of the lage local motion it is possible to use a pocedue in analogy ith estimation of the paametes of the lage global motion by using the multiple iteations [4]. hen calculation of the animation paamete fo the Fig.. Estimation of the animation paametes of the lage local motion. In the case of using featue vetices e get the linea system composed fom eq.(8) hee p êp ˆI t M = M () eˆ p p ˆI t ˆ I f v ˆ J + f h ê = I + I i y x j and Î t ae the membes fo the featue vetex,,,... Afteads by using LSM a solution of eq. () fo the animation paamete of the lage local motion ill be = ê ê ˆI t () he deived eq.() fo estimation of the animation paamete of the lage local motion is fomally the same as eq.(7) valid only fo the small local motion. In eq.() thee is futhemoe an infomation about the local motion fom the pevious fame hich enable estimation of the animation paamete of the lage local motion ith less eo. hen by application of the multiple iteations though the feedback of pediction in Fig. it is possible to incease the accuacy of estimation of the animation paamete.

6 J. MIHALÍK, V. MICHALČI, AIMAIO OF 3D MODEL OF HUMA HEAD he esult of estimation using the optical flo accoding to eq.(7) and () is the animation paamete fo AU hich the given local motion ealizes. he algoithm of estimation of the animation paametes of the local motion based on eq.(7) o () using the optical flo is vey fast because does not need extaction of the featue points in the fames. On the othe side its calculation equiements depend just on the numbe of the featue vetices. 5. Animation in combination ith global motion he local motion of the human head in the eal videosequence is often affected by its global motion and they ceate togethe the combination motion. heefoe fo animation of 3D model of the human head e ill use the designed algoithms not just fo estimation of the animation paametes but also the paametes of the global motion [4]. In addition the estimation of the global motion paametes is caied out independently on the local motion hat is done by selecting of the featue vetices ithout the local motion. On the othe side it is not possible fo estimation of the animation paametes because the selected featue vetices except fo the local motion have the global motion too. On the basis of the pevious consideation about animation in combination ith the global motion it as designed an estimato hich the block diagam is in Fig.3. Fist estimation of the global motion paametes is done fom the input videosequence. Afteads they ae used fo estimation of the animation paametes. heeof follo out that fo estimation of the animation paametes the algoithm fo the lage local motion descibed by eq.() have to be used. In this case the global motion paametes togethe ith the animation paametes fom the pevious fame ae used fo pediction of the components ( u,u ˆi ˆ j) of the vecto of optical flo. Estimation of global motion P W Coection of animation paametes Fig.3. he block diagam of an estimato of the animation paametes ith the global motion. ext e deive the equations fo estimation of the animation paametes in combination ith the global P W motion beteen to successive fames. Assume that 3D model of the human head is calibated accoding to the fist one of the to fames and its vetex M =(h,v,) in initial position in MCS. Afte animation ith global motion its ne coodinates M ' = (h',v',') ill be calculated as follos h' h h h h v' = R v + v + = R v + + v ' hee hen 3) h h = v R v (4) M ' = (h',v',') animated vetex and is the otated, tanslated and ( h, v, ) is the otated animation vecto belonging to the vetex M. Fom the definition of animation ith global motion by eq. (3) follo out that in this case the pinciple of supeposition does not valid. It is caused by the fact that the animation vecto of the vetex hich except fo the local motion has the global motion too is affected ith its otation matix R. It has to be taken into account fo estimation of the animation paametes. Afte tansfomation of the ne coodinates ( h ', v ', ') of the otated, tanslated and animated vetex M fom MCS into CCS and substitution fo the otation matix R and the tanslation vecto e get x' = x + ( Θ y +Θv( z) + th + h ) x y' = y + ( Θx Θh( z) + tv + v ) y (5) (6) z' = z + ( Θvx Θhy t z ) (7) By application of the analogical deivation pocedue as fo the global o local motion it is possible fom eq.(5), (6), (7) to deivate the linea fomulas fo the components of the vecto of optical flo + uj= Θ + Θ + f J J + f h t 0t t () () J fx h JI fx J fxi h v f y () fx fy x x + h+ v+ = + Pediction of animation paametes VP Θ + I fy v f y I JI fj y ui= + Θh Θv Θ + () fy fx fx fy I I fy v + 0t h t v+ t = + UP () () (8) (9) ext suppose that thee ae knon the global motion

RADIOEGIEERIG, VOL. 6, O., APRIL 007 63 paametes P =Θ ( h, Θv, Θ,t h,t v,t ) beteen to successive fames hich ae estimated at fist and sepaately. hen fo pediction of the components of the vecto of optical flo is valid J + fx h ûj = ŵ+ VP (30) I fy v ûi = ŵ+ UP (3) hee ŵ is the animation paamete fom the pevious fame and h, v, ae the otated components of the animation vecto ith the otation angels of the global motion (eq.4). By substitution of eq.(8) and (9) into the optical flo equation [4] of the lage motion fo one featue vetex can be obtained the folloing linea equation ith the unknon animation paamete ˆ I fy v ˆ J + fx h Ii Ij ( ˆI ˆ + = t+ Ii UP+Î j VP) (3) If featue vetices ae used then the system of linea equations ill be composed hee ep Ip = M e M p I (33) p ˆ I f v ˆ J + f h e = I + I, i y x j I = ˆI + ˆI UP+ Iˆ t i j VP ae the membes fo the featue vetex,,,... Afte application of LSM on solution of eq.(33) fo the animation paamete is valid = e I e (34) If the local motion is ealized by using moe AU then fo each of them is composed the system of eq. (33) and afteads by application of LSM on its solution the animation paamete belonging to selected AU is obtained. he deived eq.(34) is fomally the same as eq.() valid fo estimation of the animation paamete of the lage local motion. On the othe side in eq. (34) thee is futhemoe an infomation about the global motion hich paametes ae estimated independently on the animation paamete. It is clea that fo animation of 3D model Candide ith its global motion the accuacy of estimation of the animation paametes is dependent on that one of estimation of the global motion paametes befoe caied out. he accuacy can be next inceased by the multiple iteations applied on eq.(30) until (34). 6. Expeimental esults he designed algoithm of animation of 3D model of the human head ith the global motion has been expeimentally veified fo the testing videosequence MissAmeica ith the fame ate 30 Hz, the size 88x35 pels and 8 bits pe pel. As a specific 3D model e used 3D ie fame model Candide [] hich as calibated by the fist (efeence) fame of the videosequence. Fo estimation of the animation paametes thee ae too needed the camea paametes d, f x, f y, hich ee obtained by a calibation of the camea [4] accoding to the same efeence fame. Fo the distance d=400 pels the scaled focal lengths f x =354 pels and f y =333 pels and togethe ee used in all expeiments. Veifying of the accuacy of animation of 3D model of the human head is vey had because in advance thee ae not knon the exact values of the animation paametes. As a suitable citeion of valuation of the accuacy e used the peak signal/noise atio (SR) fo the human face egion. o compae the esults of animation of 3D model Candide by using the citeia in all expeiments fo its textuing as used the plane algoithm based on to dimensional affine tansfomation [3]. ote that the choice of the algoithm of textuing has not any diect impact on the accuacy of animation of 3D model Candide. hen conclusions in the pape ae valid fo any othe algoithms of textuing [5]. Fo animation of 3D model Candide in combination ith its global motion fist the paametes of the motion ae estimated ith high accuacy. It as achieved such a ay that fom 3D model Candide in initial position its 35 featue vetices have been selected as it is seen in Fig.4a. Afte thei pojection on the fame e got the featue points in hich the patial deivatives of the luminance function I(i,j,t) ee calculated by using Sobel opeato [6]. At selection of the featue vetices e poceeded unde assumption that the ones o them coesponding points on the suface of the human head in the fame ill have only the global motion. By such a pocedue it as eliminate an influence of the local motion on the accuacy of estimation of the global motion paametes. Folloing the high accuacy of the global motion estimation in the videosequence MissAmeica e used fo it the algoithm on the basis of optical flo assuming lage motion [4]. With efeence to the chaacte of the local motion in the videosequence e selected thee AU, i.e. Dopping of ja, Raising uppe lip and Dopping of eyebos fo hich the thee animation paametes W = (,, 3) ee estimated. he featue vetices of the used AU, in hich afte thei pojection on the fame the patial deivatives of the luminance function I(i,j,t) ee calculated, ae shon in Fig.4b,c,d.

64 J. MIHALÍK, V. MICHALČI, AIMAIO OF 3D MODEL OF HUMA HEAD a) b) c) d) Fig. 4. he featue vetices fo a) global motion, b) local motion of the ja, c) local motion of the uppe lip, d) local motion of the eyebos. he local motion in the videosequence ith the human head is consideed as the lage motion because in geneally it has the global motion too. In addition the numbe of the featue vetices fo AU is limited in egad to the finite numbe of its definition ones. heefoe to achieve the high accuacy of estimation of the animation paametes e used the multiple iteations in its block diagam (Fig.3). In Fig.5 thee ae the gaphs of dependences of SR of the synthesized fames afte estimation of the global motion (Glob.M) ith 3 iteations and the local motion (LM) ith,,3 and 0 iteations caused by animation of 3D model Candide using the selected thee AU. Fom the gaphs it is seen that using only one iteation by eq.(34) is not sufficient even such SR is deceased. he deceasing is caused by the big inaccuacy of estimation of the animation paametes. Bette esults e achieved aleady fo to iteations and fo moe then thee iteations the accuacy of estimation of the animation paametes W = (,, 3) does not incease. Fo illustation in Fig.6 thee ae 4 successive fames ith the pojected 3D model Candide afte its animation ith the global motion using the thee iteations. SR[dB] 3 9 7 5 3 Glob. M. LM ite. LM ite. LM 3 ite. LM 0 ite. 5 8 4 7 0 3 6 9 3 35 Fame Fig. 5. SR of the synthesized fames by animation ith the global motion. 7. Conclusion In the pape e have pesented animation of 3D model of the human head in combination ith its global motion. he designed algoithm is vey fast and ith lo calculation equiements because estimation of the animation paametes is caied out diectly fom the input videosequence. In ou expeiments e used 3D model Candide hich enables geneation of diffeent expessions by using of its AU. he estimation of the global motion paametes foegoes the estimation of the animation paametes and theefoe affects the accuacy of animation of 3D model of the human head. o incease the accuacy e have used the multiple iteations beteen to successive fames in the designed algoithm based on the optical flo. Fom the achieved esults follo out that they significantly paticulaized the estimated values of the animation paametes above all fo the lage motion. Of couse the paticulaization is at the expense of the calculation equiements but on the othe side it is necessay to suppess accumulation of the estimation eos in the long videosequences. Last but not least the calculation of the patial deivatives of the luminance function using of Sobel opeato contibuted to the accuacy and stability of animation of 3D model Candide ith the global motion. Objective valuation of the achieved esults shos on the possibility of application of the designed algoithm of animation 3D model Candide in combination ith its global motion in the model based human head coding ith vey high data compession. Regulaly fame by fame in the method only the global motion and animation paametes ae coded and tansmitted ith vey lo bit ate in the ange of 300 till 000 bits pe second. Acknoledgements he ok as suppoted by the Scientific Gant Agency of Ministy of Education and Academy of Science of Slovak epublic unde Gant o. /333/06.

RADIOEGIEERIG, VOL. 6, O., APRIL 007 65 7 8 9 0 Fig.6. a) he input fames of MissAmeica numbe 7, 8, 9, 0, b) the pojected 3D model Candide afte animation ith the global motion, c) the synthesized fames numbe 7 (SR=5dB), 8 (SR=4,99dB), 9 (SR=4,99dB), 0 (SR=3,97dB) by using the plana textuing.

66 J. MIHALÍK, V. MICHALČI, AIMAIO OF 3D MODEL OF HUMA HEAD Refeences [] he special issue of the IEEE ans.on Cicuits and Systems fo Video echnology on MPEG-4 SHC, July 004. [] SRIZIS, M., SARRIS,. (Ed.): 3D modeling and animation: Synthesis and Analysis echniques fo the Human Body. IRM Pess, Heshey, PA, July 004. [3] PADZIC, I., FORCHHEIMER, R. MPEG-4 Facial Animation: he Standad, Implementation and Aplications. John Wiley and Sons, 00. [4] MIHALÍK, J. Image Coding in Videocommunnications. Mecuy- Smekal. ISB-80-8906- 47-8, Košice, 00. (In Slovak). [5] CHAO, S., ROBISO, J. Model-Based Analysis/Synthesis Image Coding ith Eye and Mouth Patch Codebooks, In Poceedings of Vision Inteface. Banff Albeta, May 994, p. 04-09. [6] MIHALÍK, J. eual etok Clusteing Vecto Quantize Design. eual etok Wol. 993, vol., no. 5, p. 97-07. [7] ZHAG, L. Estimation of Eye and Mouth Cone Point Position in a Knoledge-Based Coding System. In Poceedings of SPIE. 996, vol. 95, p. -8. [8] YUILLE, A. L., HALLIA, P. W., COHE, D. S. Featue Extaction fom Faces Using Defomable emplates. Int. Jounal of Compute Vision. 99, vol. 8, no., p. 99-. [9] AOSZCZYSZY, P.M., HAAH, J. M., GRA, P. M. A e Appoach to Wie-Fame acking fo Semantic Model-Based Coding Moving Image Coding. Signal Pocessing: Image Communication. 000, vol. 5, p. 567-580. [0] EISER, P., GIROD, B. Analysing Facial Expessions fo Vitual Confeencing, IEEE Compute Gaphics and Applications. Septembe 998, p. 70-78. [] DAVIS, M., UCERYA, M. Coding of Facial Image Sequences by Model-Based Optical Flo, In Poceedings of the Inte. Wokshop on Synthetic-atual Hybid Coding and hee Dimensional Imaging. Rhodes (Geece), Septembe 997, p. 9-94. [] AHLBERG, J. Candide-3: An Updated Paameteised Face. Rep. o. LiH-ISY-R-36, Januay 00. [3] MIHALÍK, J., MICHALČI, V. 3D Motion Estimation and extuing of Human Head Model. Radioengineeing. 004, vol. 3, no., p. 6-3, ISS 0-5. [4] MIHALÍK, J., MICHALČI, V. 3D Motion Estimation of Human Head by Using Optical Flo. Radioengineeing. 006, vol. 5, no., p. 37-44, ISS 0-5. [5] MIHALÍK, J., MICHALČI, V. extuing of Suface of 3D Human Head Model. Radioengineeing. 004, vol. 3, no. 4, p. 44-47, ISS 0-5. [6] PRA, W. K. Digital Image Pocessing. John Wiley & Sons. e Yok, 978. About Authos... Ján MIHALÍK gaduated fom echnical Univesity in Batislava in 976. Since 979 he joined Faculty of Electical Engineeing and Infomatics of echnical Univesity of Košice, hee eceived his PhD degee in Radioelectonics in 985. Cuently, he is Full Pofesso of Electonics and elecommunications and the head of the Laboatoy of Digital Image Pocessing and Videocommunications at the Depatment of Electonics and Multimedia elecommunications. His eseach inteests include infomation theoy, image and video coding, digital image and video pocessing and multimedia videocommunications. Vikto MICHALČI as bon on 976 in Ukaine. He eceived the Ing degee fom echnical Univesity of Košice in 000. He is a PhD student at the Depatment of Electonics and Multimedia elecomunications of echnical Univesity, Košice. His eseach is focused on model based and vey lo bit ate video coding. Cuently he is oking as a develope of VRVS/EVO videoconfeencing system in Caltech-VRVS-SK team at Univesity of P. J. Šafaik in Košice.