Video-Based Face Recognition Using Probabilistic Appearance Manifolds

Size: px
Start display at page:

Download "Video-Based Face Recognition Using Probabilistic Appearance Manifolds"

Transcription

1 Video-Based Face Recogniion Using Probabilisic Appearance Manifolds Kuang-Chih Lee Jeffrey Ho Ming-Hsuan Yang David Kriegman Compuer Science Compuer Science & Engineering Honda Research Insiue Universiy of Illinois, Urbana-Champaign Universiy of California, San Diego 800 California Sree Urbana, IL La Jolla, CA Mounain View, CA Absrac This paper presens a novel mehod o model and recognize human faces in video sequences. Each regisered person is represened by a low-dimensional appearance manifold in he ambien image space. The complex nonlinear appearance manifold expressed as a collecion of subses (named pose manifolds), and he conneciviy among hem. Each pose manifold is approximaed by an affine plane. To consruc his represenaion, exemplars are sampled from videos, and hese exemplars are clusered wih a K-means algorihm; each cluser is represened as a plane compued hrough principal componen analysis (PCA). The conneciviy beween he pose manifolds encodes he ransiion probabiliy beween images in each of he pose manifold and is learned from a raining video sequences. A maximum a poseriori formulaion is presened for face recogniion in es video sequences by inegraing he likelihood ha he inpu image comes from a paricular pose manifold and he ransiion probabiliy o his pose manifold from he previous frame. To recognize faces wih parial occlusion, we inroduce a weigh mask ino he process. Exensive experimens demonsrae ha he proposed algorihm ouperforms exising frame-based face recogniion mehods wih emporal voing schemes. 1 Inroducion Face recogniion has long been an acive area of research, and numerous algorihms have been proposed over he years. However, mos research has been focused on recognizing faces from a single image. Face recogniion using video presens various challenges and opporuniies. Typically, recogniion using image sequences is done using a wo-sage sysem: a racking module and a recogniion module. Given a video frame, a racking module akes an esimae of he objec s locaion in he previous frame and reurns a subimage in he curren frame ha conains he objec. A recogniion module hen operaes on he subimage, perhaps inegraing informaion/decisions from earlier frames. In a video, head pose may vary significanly. Therefore, successful video-based face recogniion mus be able o classify faces wih a range of image plane and 3-D orienaions. In addiion, a good recogniion mehod should be robus o misalignmen errors inroduced by inaccuracies from he racking module. Meanwhile, parial occlusion poses anoher serious challenge, and his is likely o occur a some insans in unconsrained applicaions such as vision-based human compuer ineracion. On he oher hand, recogniion in video offers he opporuniy o inegrae informaion emporally across he video sequence, which may help o increase he recogniion raes. Our framework explois emporal coherence in he following ways. Firs, our proposed appearance model is composed of a collecion of pose manifolds, and a marix of ransiion probabiliies o connec hem. The ransiion probabiliies among he pose manifolds are learned from raining videos each one characerizes he probabiliy of moving from one pose o anoher pose beween any wo consecuive frames. We use he ransiion probabiliy o implicily infer he appropriae pose for each incoming video frame, and hen inegrae his informaion by Bayes rule o perform face recogniion. Therefore, our mehod effecively capures he dynamics of pose changes and hereby explois he emporal informaion in a video sequence for recogniion. Second, we use consecuive frames o define a mask whose elemens represen he probabiliy ha a pixel corresponds o an occlusion. The mask is ieraively updaed by analyzing he difference beween he observed image a each ime insance and he reconsruced image prediced from previous frame. We have implemened he proposed mehod and evaluaed i wih numerous experimens. The experimenal resuls show ha our mehod is effecive in recognizing faces in videos conaining large variaion of head moion as well as parial occlusions. This paper is organized as follows. We briefly summarize he relaed lieraure which moivaes his work in Secion 2. In Secion 3, we deail and conras our algorihms wih oher exising work. Numerous experimens on a large and raher difficul daa se are presened in Secion 4. We conclude wih remarks and fuure work in Secion 5. 1

2 2 Relaed Work Mos of he research work in he lieraure concenraes on represenaion and classificaion mehods for recognizing faces in sill and ofenimes single images [4, 24, 30]. Alhough here exis numerous face recogniion algorihms operaing on image sequences, hey ypically use emporal voing o improve idenificaion raes [12, 26, 28]. We also noe ha here exis several algorihms ha aim o exrac 2-D or 3-D face srucure from video sequences for recogniion and animaion [5, 14, 6, 7, 8, 9, 29, 11, 23]. However hese mehods require meiculous procedures o build 2-D or 3-D models, and do no fully exploi emporal informaion for recogniion. Among he few aemps aiming o ruly uilize emporal informaion for face recogniion in image sequences raher han simple voing, Li e al. presened a mehod o consruc ideniy surfaces using shape and exure models as well as kernel feaure exracion algorihms [16]. This approach esimaes pose angle firs in order o selec an appropriae shape model for racking and recogniion. However, i does no fully ake advanage of coherence informaion beween consecuive frames excep for a weighed emporal voing scheme o fi model parameers. Zhou and Chellappa [31] proposed a generic framework o rack and recognize human faces simulaneously by adding an ideniy variable o he sae vecor in he sequenial imporance sampling mehod. They hen marginalize over all sae vecors o yield an esimae of he poserior probabiliy of he ideniy variable. Though his probabilisic approach aims o inegrae moion and ideniy informaion over ime, i neverheless considers only ideniy consisency in emporal domain and hus may no work well when he arge is parially occluded. Furhermore, i is no clear how one can exend his work o deal wih large 3-D pose variaion. Krueger and Zhou [15] applied an on-line version of radial basis funcions o selec represenaive face images as exemplars from raining videos, and in urn his faciliaes racking and recogniion asks. The sae vecor in his mehod consiss of affine parameers as well as an ideniy variable, and he sae ransiion probabiliy is learned from affine ransformaions of exemplars from raining videos in a way similar o [27]. Since only 2-D affine ransformaions are considered, his model is effecive in capuring small 2-D moion bu may no deal well wih large 3-D pose variaion or occlusion. Recenly, Li e al. [17] applied piecewise linear models o capure local moion and a ransiion marix among hese models o describe nonlinear global dynamics. They applied he learned local linear models and heir dynamic ransiions o synhesize new moion video such as choreography. Our work bears some resemblance o heir mehod in he sense ha boh mehods uilize local linear models, somehing advocaed in several prior works [3, 1, 19], and boh learn he relaionships among hese models [13, 20, 21, 25]. However in his paper, we consider propagaing he probabilisic likelihood of he linear models hrough he ransiion marix (i.e., uilizing emporal informaion) o recognize human ideniy. Furhermore, we exploi he informaion learned in he local models and ransiion marix o infer missing daa in recognizing parially occluded faces. 3 Probabilisic Appearance Manifold Consider a recogniion problem wih N objecs where he images of an objec are acquired by varying he viewpoin. I is well undersood ha he se of images of an objec under varying viewing condiions can be reaed as a lowdimensional manifold in he image space as demonsraed in parameric appearance manifold work [19] or view-based Eigenspace approach [22]. The recogniion ask is sraighforward if he appearance manifold M k for each individual k is known: for a es image I, he ideniy k can be deermined by finding he manifold M k wih minimal disance o I, i.e., k = arg min d H (I, M k ). (1) k Here, d H denoes he L 2 Hausdorff disance beween he image I and M k. Le x M k denoe a poin on a manifold M k where dim(m k ) dim(i). Given a poin x M k, le he corresponding reconsruced face image be denoed Îx where dim(i) = dim(îx). If x is he poin on M k a minimal L 2 disance o I, hen d H (I, M k ) = d(i, x ) where d(, ) denoes he L 2 disance. Alernaively, x can be regarded as he resul of some nonlinear projecion of I ono M k. Ck2 Ck1 Ck3 Mk I Ck4 dh(mk,i) x Ck5 Ck6 Figure 1: Appearance manifold. A complex and nonlinear manifold can be approximaed as he union of several simpler pose manifolds; here, each pose manifold is represened by a PCA plane. Probabilisically, Equaion 1 is he resul of defining he condiional probabiliy p(k I) as p(k I) = 1 Λ exp( 1 σ 2 d2 H(I, M k )). (2) 2

3 where Λ is a normalizaion erm, and for a given image I k = arg max k p(k I). (3) In order o implemen his recogniion scheme, one mus be able o esimae he projeced poin x M k, and hen he image o model disance, d H (I, M k ), can be compued for a given I and for each M k. However, such disances can be compued accuraely only if M k is known exacly. In our case, M k is usually no known and can only be approximaed wih samples. The main par of our algorihm is o provide a probabilisic framework for esimaing x and d H (x, I). Noe ha if we define he condiional probabiliy p Mk (x I) o be he probabiliy ha among poins on M k, Î x has he smalles L 2 -disance o I, hen d H (I, M k ) = d(x, I)p Mk (x I)dx, (4) M k and Equaion 1 is equivalen o k = arg min d(x, I)p Mk (x I)dx. (5) k M k The abovemenioned formulaion shows ha d H (I, M k ) can be viewed as he expeced disance beween a single image frame I and a complex appearance manifold M k. Clearly, if M k were fully known or well-approximaed (e.g., described by some algebraic equaions), hen p Mk (x I) could be reaed as a δ funcion a he se of poins wih minimal disance o I. When sufficienly many samples are drawn from M k, he expeced disance d(i, M k ) will be a good approximaion of he rue disance. The reason is ha p Mk (x I) in he inegrand of Equaion 4 will approach a dela funcion wih is energy concenraed on he se of poins wih minimal disance o I. In our case, M k, a bes, is approximaed hrough a sparse se of samples, and so we will model p Mk (x I) wih a Gaussian disribuion. Since he appearance manifold M k is complex and nonlinear, i is reasonable o decompose M k ino a collecion of m simpler disjoin manifolds, M k = C k1 C km where C ki is called a pose manifold. Each pose manifold is furher approximaed by an affine plane compued hrough principal componen analysis (called a PCA plane). We define he condiional probabiliy p(c ki I) for 1 i m as he probabiliy ha C ki conains a poin x wih minimal disance o I. Since p Mk (x I) = m i=1 p(cki I)p C ki(x I), we have, d H (I, M k ) = d(x, I)p Mk (x I)dx M k = p(c ki I) d H (x, I)p C ki(x I)dx C ki = i=1 p(c ki I)d H (I, C ki ). (6) i=1 I-1 MA I I-2 I+1 I+2 I-3 Figure 2: Difficuly of frame-based recogniion: The wo solid curves denoe wo differen appearance manifolds, M A and M B I is difficul o reach a decision on he ideniy from frame I 3 o frame I because hese frames have smaller L 2 disance o appearance manifolds M A han M B. However, by looking a he sequence of images I 6... I +3, i is apparen ha he sequence has mos likely originaed from appearance manifold M B. The above equaion shows ha he expeced disance d(i, M k ) can be also reaed as he average expeced disance beween I and each pose manifold C ki. In addiion, his equaion ransforms he inegral o a finie summaion which is feasible o compue numerically. For face recogniion from video sequences, we can exploi emporal coherence beween consecuive image frames. As shown in Figure 2, he L 2 norm may occasionally be misleading during recogniion. Bu if we consider previous frames in an image sequence raher han jus one, hen he se of closes poins x will race a curve on a pose manifold. In our framework, his is embodied by he erm p(c ki I) in Equaion 6. In Secion 3.1, we will apply Bayesian inference o incorporae emporal informaion o provide a beer esimaion of p(c ki I), and hus d H (I, M k ) o achieve beer recogniion performance. 3.1 Compuing p(c ki I ) For recogniion from a video sequence, we need o esimae p(c ki I ) for each i a ime. To incorporae emporal informaion, p(c ki I ) should be aken as he join condiional probabiliy p(c ki I, I 0: 1 ) where I 0: 1 denoes he frames from he beginning up o ime 1. We furher assume I and I 0: 1 are independen given C ki, as well as C ki and I 0: 1 are independen given C 1. ki Using Bayes rule we have he following recursive formulaion: p(c ki I+3 I-4 I, I 0: 1 ) = α p(i C ki = α p(i C ki ) = α p(i C ki ) j=1 j=1 MB I-5 I-6, I 0: 1 )p(c ki I 0: 1 ) p(c ki C kj 1, I 0: 1)p(C kj 1 I 0: 1) p(c ki C kj 1 )p(ckj 1 I 1I 0: 2 )(7) 3

4 where α is a normalizaion erm o ensure a proper probabiliy disribuion. The emporal dynamics of he video sequence is capured by he ransiion probabiliy beween he manifolds, p(c ki C kj 1 ). Noe ha p(cki C kj 1 C kj 1 ) is he probabiliy of x C ki given x 1 C kj. For wo consecuive frames I 1 and I, because of emporal coherency, we expec ha heir projeced poins x 1 and x should have small geodesic disance on M (See Figure 2). Tha is he ransiion probabiliy p(c ki geodesic disance beween C ki and C kj. Ck1 P(Ck1 Ck2) Ck2 Mk ) is relaed implicily o he P(Ck2 Ck3) Ck3 Figure 3: Dynamics among pose manifolds. The dynamics among he pose manifolds are learned from raining videos which describes he probabiliy of moving from one manifold o anoher a any ime insance. 3.2 Learning Manifolds and Dynamics For each person k, we collec a leas one video sequence conaining l consecuive images S k = {I 1,, I l }. We furher assume ha each raining image is a fair sample drawn from he appearance manifold M k. There are hree seps in he algorihm. We firs pariion hese samples ino m disjoin subses {S 1,, S m }. For each collecion S ki, we can consider i as conaining poins drawn from some pose manifold C ki of M k, and from he images in S ki, we consruc a linear approximaion o he C ki of he rue manifold M k. Afer all he C ki have been compued, we esimae he ransiion probabiliies p(c ki C kj ) for i j. In he firs sep, we apply a K-means clusering algorihm o he se of images in he video sequences. We iniialize m seeds by finding m frames from he raining videos wih he larges L 2 disance o each oher. Then he general K-means algorihm is used o assign images o he m clusers. As our goal in performing clusering is o approximae he daa se raher han o derive semanically meaningful cluser ceners, i is worh noing ha he resuling clusers are no worse han wice wha he opimal cener would be if hey could be easily found [10]. Second, for each S ki we obain a linear approximaion of he underlying subse C ki M k by compuing a PCA plane L ki of fixed dimension for he images in S ki. Since he PCA planes approximae appearance manifold M i, heir dimension is he inrinsic dimension of M, and herefore all PCA planes L i have he same dimension. Finally, he ransiion probabiliy p(c ki C kj ) is defined by couning he acual ransiions beween differen S i observed in he image sequence: p(c ki C kj ) = 1 Λ ki l δ(i q 1 S ki )δ(i q S kj ) (8) q=2 where δ(i q S kj ) = 1 if I q S kj and oherwise i is 0. The normalizing consan Λ ki ensures ha p(c ki C kj ) = 1. (9) j=1 where we se p(c ki C ki ) o a consan κ. A graphic represenaion of a ransiion marix wih m = 5 learned from a raining video is depiced in Figure 4. Wih C ki and is linear approximaion L ki defined, we can define how p(i C ki ) can be calculaed. We can compue he L 2 disances ˆd ki = d H (I, L ki ) from I o each L ki. We rea ˆd ki as an esimae of he rue disance from I o C ki, i.e., d H (I, C ki ) = d H (I, L ki ). p(i C ki ) is defined as p(i C ki ) = 1 Λ ki exp( 1 2 σ 2 ˆd 2 ki) (10) wih Λ ki = m 1 i=1 exp( ˆd 2 2 σ 2 ki ). Noice ha we use a non-compac subspace L ki o approximae a compac pose manifold C ki. The infinie exen of L ki migh be beer capured by he underlying Gaussian, and similar work has been done by Moghaddam e al.[18]. However, our experimen shows ha he recogniion resul using his more elaborae algorihm is no beer han he one proposed in he paper. This can be explained by he fac ha alhough he linear subspaces are non-compac, he es images will almos always be drawn from a compac subse of he image space. This effec makes he subspaces funcionally compac in our algorihm. In oher words, he subspaces behave as hey only have finie exen. 3.3 Face Recogniion from Video Given an image I from a video sequence, we compue for each person k he disance d H (I, M k ) using he Equaion 6. Noe ha p(c ki I) has a emporal dependency, and i is compued recursively using Equaion 7. Once all he d H (I, M k ) have been compued, he poserior p(k I) is compued by Equaion 2 wih appropriae σ, and he human ideniy is decided by Equaion 5. 4

5 Pose Figure 4: Graphic represenaion of a ransiion marix learned from a raining video. In his example, he appearance manifold is approximaed by 5 pose subspaces. The reconsruced cener image of each pose subspace is shown a he op raw and column. The ransiion probabiliy marix is drawn by he 5 5 block diagram. The brigher block means a higher ransiion probabiliy. I is easy o see ha he fronal pose (pose 1) has higher probabiliy o change o oher poses; he righ pose (pose 2) has almos zero probabiliy o direcly change o he lef pose (pose 3). when compuing d H (M k, I ). We inroduce an image mask W, which defines he probabiliy ha a pixel is occluded, where W has he same dimension as image I, and is elemens are iniialized wih a 1, i.e., assuming here is no occlusion a he firs frame and no pixel is downweighed. The d H (M k, I ) is hen replaced by he weighed disance d H (M k, W. I ) where. denoe elemen-by-elemen muliplicaion. Le he weighed projecion of W. I on M k be x, he mask W is updaed in each frame I by he esimae a a previous frame W 1 by W (1) = exp( 1 2 σ 2 (Îx I ). (Îx I )) (11) in he firs ieraion. Alernaively, W can be ieraively updaed based on he W (1) and Î(1) x (i.e., he reconsruced image based on W (1) and d H (M k, W (1). I )) W (i+1) = exp( 1 2 σ 2 (Î(i) x I ). (Î(i) x I )) (12) unil he difference beween W (i) and W (i 1) hreshold value a he i-h ieraion. is below a I is also worh menioning ha he proposed framework explois he emporal coherence in he appearance of consecuive face images by inegraing he manifold ransiion a he previous and curren ime insance. For face recogniion wih varying pose, our mehod ensures ha he ransiions beween pose manifolds do no occur arbirarily bu raher in a consrained order. For example he appearance of one person s face canno change immediaely from lef profile o righ profile in wo consecuive frames, bu raher i mus pass hrough some inermediae pose or orienaion (See Figure 6). This process can also be considered as puing a firs order Markov process or finie sae machine over a piecewise linear srucure. In conras, simple emporal voing scheme has been commonly adoped in mos video-based face recogniion mehods [16] [26]. 3.4 Recognizing Parially Occluded Faces Similar o our formulaion exploiing emporal informaion for recogniion, he same approach can be easily exended o deal wih parial occlusion of a face by considering he previous frame as prior informaion. The original formulaion for d H (C ki, I ) reas every pixel in image I wih equal weigh assuming ha here is no occlusion anywhere in he image sequence. If we knew which pixels corresponded o occlusions, we would pu lower weighs on hose pixels Figure 5: Top row: (lef) an unoccluded face image, (cener) a reconsruced image using corresponding pose manifold, and (righ) a corresponding mask). Boom row: (lef) a face image parially occluded by one hand, (cener) a reconsruced image using corresponding pose manifold, and (righ) an updaed mask. Boh he appearance manifold and mask informaion a previous frames are uilized o esimae he curren occlusion mask in he equaions above. We firs perform he weighed projecion o find a reconsruced image using he corresponding pose manifold and ieraively esimae he occlusion areas in he curren frame. Once we ge an updaed mask W in frame I by Equaion 11, we evaluae Equaion 6 for face recogniion by replacing d H (C ki, I ) wih d H (C ki, W. I ). Figure 5 shows an example where a face is parially occluded by an objec (lower lef). The reconsruced image using he corresponding pose manifold is shown in he lower cener. The updaed mask is shown in he lower righ where he values have been hresholded a dark pixel denoes a probabiliy of occlusion. Noe ha he updaed mask maches he occluded region reasonably well. Noe also ha 5

6 he mask predics ha several pixels are occluded hough in fac hey are no. This is caused by he disagreemen beween he inpu image and he reconsruced image. Neverheless, he regions ha maer mos for recogniion (i.e., he cenral face region and he occluded region) are weighed appropriaely. Our experimenal resuls, presened in he nex secion, also demonsrae ha he mask scheme is effecive in recognizing parially occluded faces. 4 Experimens and Resuls We evaluaed he proposed algorihm on wo ses of videos: one wihou any occlusion and one wih parial occlusion. The overall recogniion rae in he experimens is defined by he number frames where he ideniy is correcly recognized divided by he number of frames in all he es videos. 4.1 Number of Linear PCA Planes Recogniion Rae (%) Number of PCA Planes Figure 6: Sample gallery videos used in he experimens. Noe he pose variaion changed is raher large in his daa se. We performed numerous experimens and compared he proposed algorihm wih oher mehods in he conex of video-based recogniion. Since here is no sandard daabase ha conains large 2-D and 3-D head roaion for video-based face recogniion, we colleced a se of 45 videos of 20 differen people for experimens (This daa se will be made available o he vision communiy in he near fuure.). Each individual in our daabase has a leas wo videos where each person moves in a differen combinaion of 2-D and 3-D roaion, expression, and speed. Each video was recorded in an indoor environmen and each one lased for a leas 20 seconds (wih 30 color frames of pixels per second). Some cropped frames from he videos are shown in Figure 6. A varian of he eigen-subspace racker [2] was used o locae he face, and he resuls were inspeced by humans. Each image was hen downsampled o pixels for compuaional efficiency. To reduce he effec of misalignmen caused by he racker, we added small 2-D perurbaions including ranslaion (wihin 2 pixels in all direcions), and scaling (wihin a scale from 0.9 o 1.1), o enlarge he raining ses before applying he proposed probabilisic algorihm. Figure 7: Recogniion rae vs. number of piecewise linear PCA planes of our mehod. I shows ha he proposed mehod is raher robus o parameer selecion (i.e., he number of pose manifolds used in approximaing appearance manifold.) We firs evaluae he proposed algorihm in he es se wihou occlusion, and analyze he number of PCA planes required o consruc appearance manifolds yielding good recogniion resuls. Figure 7 demonsrae ha he average recogniion rae does no change much when he number of PCA planes is varied from 5 o 30. The resuls sugges ha he appearance manifold can be effecively approximaed wih a small number of PCA planes. The proposed algorihm performs well over a reasonably large range which shows ha one can easily pick an appropriae number of PCA planes. Obviously, a smaller number of PCA planes is preferable for compuaional efficiency reasons. However, he recogniion rae drops significanly and quickly when he number of manifolds is raher small (fewer han five for his daa se). This is consisen wih he claim ha he appearance manifold is nonlinear and complex. 4.2 Transiion Marix P (C ki C kj ) In his se of experimens, we demonsrae ha he ransiion marix, P (C ki C kj ), in he proposed mehod capure he image dynamics sufficienly o improve recogni- 6

7 COMPARISON OF TEMPORAL STRATEGIES Temporal Sraegy Accuracy (%) Proposed Mehod 92.1 Temporal Voing 84.2 Uniform Trans Table 1: Recogniion resuls using various emporal sraegies on a es se of videos wihou occlusion. ion raes. Using he se of videos wihou occlusion, we compared our mehod wih wo differen sraegies, emporal voing and a uniform ransiion probabiliy scheme. All hree mehods used he same number of manifolds for each person m = 5; hey differ in heir way of uilizing emporal informaion. The emporal voing scheme, commonly used in recogniion mehods is based on muliple frames, makes an ideniy decision by aking voes of he resuls of he previous f frames. In his case, 20 frames were used. The uniform ransiion scheme simply ses all he enries of ransiion marix o 1, which means ha no emporal dynamics are learned or uilized in he recogniion process. The experimenal resuls, shown in Table 1, demonsrae ha our mehod ouperforms ohers by a significan margin. In oher words, learning ransiion probabiliies among he pose manifolds does faciliae recogniion which canno be achieved by mehod using no dynamics informaion or a simple emporal voing scheme wih a large window size. 4.3 Comparison wih Single Frame Algorihms and he Effec of Occlusion COMPARISON OF RECOGNITION METHODS Mehod Accuracy (%) Videos w/o Videos wih occlusion occlusion Proposed Mehod Ensemble of LPCA Eigenface Fisherface Table 2: Recogniion resuls using differen mehods. The resuls are based on he average recogniion raes achieved by each mehod. For compleeness, we compared our mehod wih several frame-based face recogniion algorihms in he lieraure, and he resuls are shown in Table 2. All mehods were rained wih he exac same cropped images. We consruced 30 PCA planes and learn heir dynamics from he raining videos in he proposed algorihm. For he Ensemble of LPCA mehod, we used he same 30 PCA planes consruced in he proposed mehod bu did no use he learned ransiion marix. This mehod is, in spiri, similar o he view-based Eigenface mehod [22]. The dimensionaliy of Fisherface mehod is se o 19 (i.e., he number of classes minus 1) and he dimensionaliy for oher mehods is empirically se o 30. Though i may no seem o be fair o compare video-based and frame-based recogniion algorihms, hese baseline experimens sugges ha frame-based mehods may no work well in an unconsrained environmen where here are large pose changes. For he es videos wihou occlusion, he Ensemble of LPCA mehod performs beer han classic linear models (Eigenface and Fisherface mehods) because an image sequence usually conain 2-D and 3-D roaions, which can no be effecively approximaed by a global linear model. These resuls also show ha he use of image dynamics by our mehod grealy helps face recogniion in video. Excep for he proposed mehod, all oher mehods performed poorly on he es videos where some faces were parially occluded. This resul shows ha appearance coherence beween consecuive frames helps in predicing occlusions and in urn faciliaes he recogniion process. 5 Conclusion and Fuure Work We have presened a novel framework for video-based face recogniion. The proposed mehod builds an appearance manifold which is approximaed by piecewise linear subspaces and he dynamics among hem embodied in a ransiion marix learned from an image sequence. I is worh noicing ha he image sequences considered in his paper conains large 2-D and 3-D roaions as well as parial occlusions. These siuaions migh occur in many visionbased human-compuer ineracion or surveillance applicaions. As experimenally demonsraed, our mehod approximaes nonlinear appearance manifold well and achieves good recogniion raes in video-based face recogniion. Though he proposed model handles large moions well, i is neverheless sensiive o large illuminaion changes, and our fuure work will address his. Acknowledgmens Suppor of his work was provided by Honda Research Insiue, and he Naional Science Foundaion CCR and IIS This work was carried ou a Honda Research Insiue. We would like o hank he anonymous reviewers for heir commens and suggesions, and all he people who help o record heir faces in our video daabase. 7

8 References [1] C. M. Bishop and J. M. Winn. Non-linear Bayesian image modelling. In Proc. European Conf. on Compuer Vision, volume 1, pages 3 17, [2] M. J. Black and A. D. Jepson. Eigenracking: Robus maching and racking of ariculaed objecs using a view-based represenaion. In l. J. Compuer Vision, 26(1):63 84, [3] C. Bregler and S. Omohundro. Surface learning wih applicaions o lipreading. In Advances in Neural Informaion Processing Sysems, pages 43 50, [4] R. Chellappa, C. L. Wilson, and S. Sirohey. Human and machine recogniion of faces: A survey. Proceedings of he IEEE, 83(5): , [5] T. Cooes, C. J. Taylor, D. Cooper, and J. Graham. Acive shape models - Their raining and applicaion. Compuer Vision and Image Undersanding, 61:38 59, [6] D. DeCarlo, D. Meaxas, and M. Sone. An anhropomeric face model using variaional echniques. In Proc. SIG- GRAPH, pages 67 74, [7] G. J. Edwards, C. J. Taylor, and T. F. Cooes. Inerpreing face images using acive appearance models. In Proc. IEEE In l. Conf. on Auomaic Face and Gesure Recogniion, pages , [8] G. J. Edwards, C. J. Taylor, and T. F. Cooes. Improving idenificaion performance by inegraing evidence from sequence. In Proc. IEEE Conf. on Compuer Vision and Paern Recogniion, pages , [9] A. S. Georghiades, P. N. Belhumeur, and D. J. Kriegman. From few o many: Illuminaion cone models for face recogniion under variable lighing and pose. IEEE Trans. Paern Analysis and Machine Inelligence, 23(6): , [10] D. Hochbaum and D. Shmoys. A bes possible heurisic for he k-cener problem. Mahemaics of Operaions Research, 10: , [11] X. Hou, S. Li, H. Zhang, and Q. Cheng. Direc appearance models. In Proc. IEEE Conf. on Compuer Vision and Paern Recogniion, volume 1, pages , [12] A. J. Howell and H. Buxon. Towards unconsrained face recogniion from image sequences. In Proc. IEEE In l. Conf. on Auomaic Face and Gesure Recogniion, pages , [13] M. Isard and A. Blake. A mixed-sae Condensaion racker wih auomaic model-swiching. pages , [14] T. Jebara, K. Russell, and A. Penland. Mixures of eigen feaures for real-ime srucure from exure. In Proc. In l. Conf. on Compuer Vision, pages , [15] V. Krüeger and S. Zhou. Exemplar-based face recogniion from video. In Proc. European Conf. on Compuer Vision, volume 4, pages [16] Y. Li, S. Gong, and H. Liddell. Consrucing facial ideniy surface in a nonlinear discriminaing space. In Proc. IEEE Conf. on Compuer Vision and Paern Recogniion, volume 2, pages , [17] Y. Li, T. Wang, and H.-Y. Shum. Moion exures: A wolevel saisical model for characer moion synhesis. In Proc. SIGGRAPH, pages , [18] B. Moghaddam and A. Penland. Probabilisic visual learning for objec recogniion. IEEE Trans. Paern Analysis and Machine Inelligence, 19(7): , [19] H. Murase and S. K. Nayar. Visual learning and recogniion of 3-D objecs from appearance. In l. J. Compuer Vision, 14:5 24, [20] B. Norh, A. Blake, M. Isard, and J. Rischer. Learning and classificaion of complex dynamics. IEEE Trans. Paern Analysis and Machine Inelligence, 22(9): , [21] V. Pavlović, J. M. Rehg, T. J. Cham, and K. P. Murphy. A dynamic Bayesian nework approach o figure racking using learned dynamic models. In Proc. In l. Conf. on Compuer Vision, pages , [22] A. Penland, B. Moghaddam, and T. Sarner. View-based and modular eigenspaces for face recogniion. In Proc. IEEE Conf. on Compuer Vision and Paern Recogniion, [23] S. Romdhani, V. Blanz, and T. Veer. Face idenificaion by fiing 3D morphable model using linear shape and exure error funcions. pages 3 19, [24] A. Samal and P. A. Iyengar. Auomaic recogniion and analysis of human faces and facial expressions: A survey. Paern Recogniion, 25(1):65 77, [25] A. Schödl, R. Szeliski, D. H. Salesin, and I. Essa. Video exures. In Proc. SIGGRAPH, pages , [26] G. Shakhnarovich, J. W. Fisher, and T. Darrell. Face recogniion from long-erm observaions. In Proc. European Conf. on Compuer Vision, volume 3, pages , [27] K. Toyama and A. Blake. Probabilisic racking in a meric space. In Proc. In l. Conf. on Compuer Vision, volume 2, pages 50 59, [28] H. Wechsler, V. Kakkad, J. Huang, S. Gua, and V. Chen. Auomaic video-based person auhenicaion using he RBF nework. In Proc. In l. Conf. on Audio and Video-Based Biomeric Person Auhenicaion, pages , [29] W. Y. Zhao and R. Chellappa. Symmeric shape-fromshading using self-raio image. In l. J. Compuer Vision, 45(1):55 75, [30] W. Y. Zhao, R. Chellappa, A. Rosenfeld, and J. P. Phillips. Face recogniion: A lieraure survey. Technical Repor CAR-TR-948, Cener for Auomaion Research, Universiy of Maryland, [31] S. Zhou and R. Chellappa. Probabilisic human recogniion from video. In Proc. European Conf. on Compuer Vision, volume 3, pages ,

Implementing Ray Casting in Tetrahedral Meshes with Programmable Graphics Hardware (Technical Report)

Implementing Ray Casting in Tetrahedral Meshes with Programmable Graphics Hardware (Technical Report) Implemening Ray Casing in Terahedral Meshes wih Programmable Graphics Hardware (Technical Repor) Marin Kraus, Thomas Erl March 28, 2002 1 Inroducion Alhough cell-projecion, e.g., [3, 2], and resampling,

More information

STEREO PLANE MATCHING TECHNIQUE

STEREO PLANE MATCHING TECHNIQUE STEREO PLANE MATCHING TECHNIQUE Commission III KEY WORDS: Sereo Maching, Surface Modeling, Projecive Transformaion, Homography ABSTRACT: This paper presens a new ype of sereo maching algorihm called Sereo

More information

A Matching Algorithm for Content-Based Image Retrieval

A Matching Algorithm for Content-Based Image Retrieval A Maching Algorihm for Conen-Based Image Rerieval Sue J. Cho Deparmen of Compuer Science Seoul Naional Universiy Seoul, Korea Absrac Conen-based image rerieval sysem rerieves an image from a daabase using

More information

CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL

CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL Klečka Jan Docoral Degree Programme (1), FEEC BUT E-mail: xkleck01@sud.feec.vubr.cz Supervised by: Horák Karel E-mail: horak@feec.vubr.cz

More information

Learning in Games via Opponent Strategy Estimation and Policy Search

Learning in Games via Opponent Strategy Estimation and Policy Search Learning in Games via Opponen Sraegy Esimaion and Policy Search Yavar Naddaf Deparmen of Compuer Science Universiy of Briish Columbia Vancouver, BC yavar@naddaf.name Nando de Freias (Supervisor) Deparmen

More information

Visual Perception as Bayesian Inference. David J Fleet. University of Toronto

Visual Perception as Bayesian Inference. David J Fleet. University of Toronto Visual Percepion as Bayesian Inference David J Flee Universiy of Torono Basic rules of probabiliy sum rule (for muually exclusive a ): produc rule (condiioning): independence (def n ): Bayes rule: marginalizaion:

More information

Tracking Appearances with Occlusions

Tracking Appearances with Occlusions Tracking ppearances wih Occlusions Ying Wu, Ting Yu, Gang Hua Deparmen of Elecrical & Compuer Engineering Norhwesern Universiy 2145 Sheridan oad, Evanson, IL 60208 {yingwu,ingyu,ganghua}@ece.nwu.edu bsrac

More information

Image segmentation. Motivation. Objective. Definitions. A classification of segmentation techniques. Assumptions for thresholding

Image segmentation. Motivation. Objective. Definitions. A classification of segmentation techniques. Assumptions for thresholding Moivaion Image segmenaion Which pixels belong o he same objec in an image/video sequence? (spaial segmenaion) Which frames belong o he same video sho? (emporal segmenaion) Which frames belong o he same

More information

Probabilistic Detection and Tracking of Motion Discontinuities

Probabilistic Detection and Tracking of Motion Discontinuities Probabilisic Deecion and Tracking of Moion Disconinuiies Michael J. Black David J. Flee Xerox Palo Alo Research Cener 3333 Coyoe Hill Road Palo Alo, CA 94304 fblack,fleeg@parc.xerox.com hp://www.parc.xerox.com/fblack,fleeg/

More information

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS PART A: SYSTEMS AND HUMANS 1

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS PART A: SYSTEMS AND HUMANS 1 TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS PART A: SYSTEMS AND HUMANS 1 Adapive Appearance Model and Condensaion Algorihm for Robus Face Tracking Yui Man Lui, Suden Member,, J. Ross Beveridge, Member,,

More information

Gauss-Jordan Algorithm

Gauss-Jordan Algorithm Gauss-Jordan Algorihm The Gauss-Jordan algorihm is a sep by sep procedure for solving a sysem of linear equaions which may conain any number of variables and any number of equaions. The algorihm is carried

More information

EECS 487: Interactive Computer Graphics

EECS 487: Interactive Computer Graphics EECS 487: Ineracive Compuer Graphics Lecure 7: B-splines curves Raional Bézier and NURBS Cubic Splines A represenaion of cubic spline consiss of: four conrol poins (why four?) hese are compleely user specified

More information

J. Vis. Commun. Image R.

J. Vis. Commun. Image R. J. Vis. Commun. Image R. 20 (2009) 9 27 Conens liss available a ScienceDirec J. Vis. Commun. Image R. journal homepage: www.elsevier.com/locae/jvci Face deecion and racking using a Boosed Adapive Paricle

More information

Occlusion-Free Hand Motion Tracking by Multiple Cameras and Particle Filtering with Prediction

Occlusion-Free Hand Motion Tracking by Multiple Cameras and Particle Filtering with Prediction 58 IJCSNS Inernaional Journal of Compuer Science and Nework Securiy, VOL.6 No.10, Ocober 006 Occlusion-Free Hand Moion Tracking by Muliple Cameras and Paricle Filering wih Predicion Makoo Kao, and Gang

More information

Coded Caching with Multiple File Requests

Coded Caching with Multiple File Requests Coded Caching wih Muliple File Requess Yi-Peng Wei Sennur Ulukus Deparmen of Elecrical and Compuer Engineering Universiy of Maryland College Park, MD 20742 ypwei@umd.edu ulukus@umd.edu Absrac We sudy a

More information

CENG 477 Introduction to Computer Graphics. Modeling Transformations

CENG 477 Introduction to Computer Graphics. Modeling Transformations CENG 477 Inroducion o Compuer Graphics Modeling Transformaions Modeling Transformaions Model coordinaes o World coordinaes: Model coordinaes: All shapes wih heir local coordinaes and sies. world World

More information

NEWTON S SECOND LAW OF MOTION

NEWTON S SECOND LAW OF MOTION Course and Secion Dae Names NEWTON S SECOND LAW OF MOTION The acceleraion of an objec is defined as he rae of change of elociy. If he elociy changes by an amoun in a ime, hen he aerage acceleraion during

More information

Sam knows that his MP3 player has 40% of its battery life left and that the battery charges by an additional 12 percentage points every 15 minutes.

Sam knows that his MP3 player has 40% of its battery life left and that the battery charges by an additional 12 percentage points every 15 minutes. 8.F Baery Charging Task Sam wans o ake his MP3 player and his video game player on a car rip. An hour before hey plan o leave, he realized ha he forgo o charge he baeries las nigh. A ha poin, he plugged

More information

Improved TLD Algorithm for Face Tracking

Improved TLD Algorithm for Face Tracking Absrac Improved TLD Algorihm for Face Tracking Huimin Li a, Chaojing Yu b and Jing Chen c Chongqing Universiy of Poss and Telecommunicaions, Chongqing 400065, China a li.huimin666@163.com, b 15023299065@163.com,

More information

FACIAL ACTION TRACKING USING PARTICLE FILTERS AND ACTIVE APPEARANCE MODELS. Soumya Hamlaoui & Franck Davoine

FACIAL ACTION TRACKING USING PARTICLE FILTERS AND ACTIVE APPEARANCE MODELS. Soumya Hamlaoui & Franck Davoine FACIAL ACTION TRACKING USING PARTICLE FILTERS AND ACTIVE APPEARANCE MODELS Soumya Hamlaoui & Franck Davoine HEUDIASYC Mixed Research Uni, CNRS / Compiègne Universiy of Technology BP 20529, 60205 Compiègne

More information

Real time 3D face and facial feature tracking

Real time 3D face and facial feature tracking J Real-Time Image Proc (2007) 2:35 44 DOI 10.1007/s11554-007-0032-2 ORIGINAL RESEARCH PAPER Real ime 3D face and facial feaure racking Fadi Dornaika Æ Javier Orozco Received: 23 November 2006 / Acceped:

More information

In Proceedings of CVPR '96. Structure and Motion of Curved 3D Objects from. using these methods [12].

In Proceedings of CVPR '96. Structure and Motion of Curved 3D Objects from. using these methods [12]. In Proceedings of CVPR '96 Srucure and Moion of Curved 3D Objecs from Monocular Silhouees B Vijayakumar David J Kriegman Dep of Elecrical Engineering Yale Universiy New Haven, CT 652-8267 Jean Ponce Compuer

More information

4. Minimax and planning problems

4. Minimax and planning problems CS/ECE/ISyE 524 Inroducion o Opimizaion Spring 2017 18 4. Minima and planning problems ˆ Opimizing piecewise linear funcions ˆ Minima problems ˆ Eample: Chebyshev cener ˆ Muli-period planning problems

More information

Robust parameterized component analysis: theory and applications to 2D facial appearance models

Robust parameterized component analysis: theory and applications to 2D facial appearance models Compuer Vision and Image Undersanding 91 (2003) 53 71 www.elsevier.com/locae/cviu Robus parameerized componen analysis: heory and applicaions o 2D facial appearance models Fernando De la Torre a, * and

More information

Multiple View Discriminative Appearance Modeling with IMCMC for Distributed Tracking

Multiple View Discriminative Appearance Modeling with IMCMC for Distributed Tracking Muliple View Discriminaive ing wih IMCMC for Disribued Tracking Sanhoshkumar Sunderrajan, B.S. Manjunah Deparmen of Elecrical and Compuer Engineering Universiy of California, Sana Barbara {sanhosh,manj}@ece.ucsb.edu

More information

Robust 3D Visual Tracking Using Particle Filtering on the SE(3) Group

Robust 3D Visual Tracking Using Particle Filtering on the SE(3) Group Robus 3D Visual Tracking Using Paricle Filering on he SE(3) Group Changhyun Choi and Henrik I. Chrisensen Roboics & Inelligen Machines, College of Compuing Georgia Insiue of Technology Alana, GA 3332,

More information

Visual Indoor Localization with a Floor-Plan Map

Visual Indoor Localization with a Floor-Plan Map Visual Indoor Localizaion wih a Floor-Plan Map Hang Chu Dep. of ECE Cornell Universiy Ihaca, NY 14850 hc772@cornell.edu Absrac In his repor, a indoor localizaion mehod is presened. The mehod akes firsperson

More information

Joint Feature Learning With Robust Local Ternary Pattern for Face Recognition

Joint Feature Learning With Robust Local Ternary Pattern for Face Recognition Join Feaure Learning Wih Robus Local Ternary Paern for Face Recogniion Yuvaraju.M 1, Shalini.S 1 Assisan Professor, Deparmen of Elecrical and Elecronics Engineering, Anna Universiy Regional Campus, Coimbaore,

More information

Robust Multi-view Face Detection Using Error Correcting Output Codes

Robust Multi-view Face Detection Using Error Correcting Output Codes Robus Muli-view Face Deecion Using Error Correcing Oupu Codes Hongming Zhang,2, Wen GaoP P, Xilin Chen 2, Shiguang Shan 2, and Debin Zhao Deparmen of Compuer Science and Engineering, Harbin Insiue of Technolog

More information

A Hierarchical Object Recognition System Based on Multi-scale Principal Curvature Regions

A Hierarchical Object Recognition System Based on Multi-scale Principal Curvature Regions A Hierarchical Objec Recogniion Sysem Based on Muli-scale Principal Curvaure Regions Wei Zhang, Hongli Deng, Thomas G Dieerich and Eric N Morensen School of Elecrical Engineering and Compuer Science Oregon

More information

Image Content Representation

Image Content Representation Image Conen Represenaion Represenaion for curves and shapes regions relaionships beween regions E.G.M. Perakis Image Represenaion & Recogniion 1 Reliable Represenaion Uniqueness: mus uniquely specify an

More information

Rao-Blackwellized Particle Filtering for Probing-Based 6-DOF Localization in Robotic Assembly

Rao-Blackwellized Particle Filtering for Probing-Based 6-DOF Localization in Robotic Assembly MITSUBISHI ELECTRIC RESEARCH LABORATORIES hp://www.merl.com Rao-Blackwellized Paricle Filering for Probing-Based 6-DOF Localizaion in Roboic Assembly Yuichi Taguchi, Tim Marks, Haruhisa Okuda TR1-8 June

More information

Upper Body Tracking for Human-Machine Interaction with a Moving Camera

Upper Body Tracking for Human-Machine Interaction with a Moving Camera The 2009 IEEE/RSJ Inernaional Conference on Inelligen Robos and Sysems Ocober -5, 2009 S. Louis, USA Upper Body Tracking for Human-Machine Ineracion wih a Moving Camera Yi-Ru Chen, Cheng-Ming Huang, and

More information

Robust LSTM-Autoencoders for Face De-Occlusion in the Wild

Robust LSTM-Autoencoders for Face De-Occlusion in the Wild IEEE TRANSACTIONS ON IMAGE PROCESSING, DRAFT 1 Robus LSTM-Auoencoders for Face De-Occlusion in he Wild Fang Zhao, Jiashi Feng, Jian Zhao, Wenhan Yang, Shuicheng Yan arxiv:1612.08534v1 [cs.cv] 27 Dec 2016

More information

A Face Detection Method Based on Skin Color Model

A Face Detection Method Based on Skin Color Model A Face Deecion Mehod Based on Skin Color Model Dazhi Zhang Boying Wu Jiebao Sun Qinglei Liao Deparmen of Mahemaics Harbin Insiue of Technology Harbin China 150000 Zhang_dz@163.com mahwby@hi.edu.cn sunjiebao@om.com

More information

A Bayesian Approach to Video Object Segmentation via Merging 3D Watershed Volumes

A Bayesian Approach to Video Object Segmentation via Merging 3D Watershed Volumes A Bayesian Approach o Video Objec Segmenaion via Merging 3D Waershed Volumes Yu-Pao Tsai 1,3, Chih-Chuan Lai 1,2, Yi-Ping Hung 1,2, and Zen-Chung Shih 3 1 Insiue of Informaion Science, Academia Sinica,

More information

Robust Visual Tracking for Multiple Targets

Robust Visual Tracking for Multiple Targets Robus Visual Tracking for Muliple Targes Yizheng Cai, Nando de Freias, and James J. Lile Universiy of Briish Columbia, Vancouver, B.C., Canada, V6T 1Z4 {yizhengc, nando, lile}@cs.ubc.ca Absrac. We address

More information

Mobile Robots Mapping

Mobile Robots Mapping Mobile Robos Mapping 1 Roboics is Easy conrol behavior percepion modelling domain model environmen model informaion exracion raw daa planning ask cogniion reasoning pah planning navigaion pah execuion

More information

COSC 3213: Computer Networks I Chapter 6 Handout # 7

COSC 3213: Computer Networks I Chapter 6 Handout # 7 COSC 3213: Compuer Neworks I Chaper 6 Handou # 7 Insrucor: Dr. Marvin Mandelbaum Deparmen of Compuer Science York Universiy F05 Secion A Medium Access Conrol (MAC) Topics: 1. Muliple Access Communicaions:

More information

Real-time 2D Video/3D LiDAR Registration

Real-time 2D Video/3D LiDAR Registration Real-ime 2D Video/3D LiDAR Regisraion C. Bodenseiner Fraunhofer IOSB chrisoph.bodenseiner@iosb.fraunhofer.de M. Arens Fraunhofer IOSB michael.arens@iosb.fraunhofer.de Absrac Progress in LiDAR scanning

More information

Reinforcement Learning by Policy Improvement. Making Use of Experiences of The Other Tasks. Hajime Kimura and Shigenobu Kobayashi

Reinforcement Learning by Policy Improvement. Making Use of Experiences of The Other Tasks. Hajime Kimura and Shigenobu Kobayashi Reinforcemen Learning by Policy Improvemen Making Use of Experiences of The Oher Tasks Hajime Kimura and Shigenobu Kobayashi Tokyo Insiue of Technology, JAPAN genfe.dis.iech.ac.jp, kobayasidis.iech.ac.jp

More information

4.1 3D GEOMETRIC TRANSFORMATIONS

4.1 3D GEOMETRIC TRANSFORMATIONS MODULE IV MCA - 3 COMPUTER GRAPHICS ADMN 29- Dep. of Compuer Science And Applicaions, SJCET, Palai 94 4. 3D GEOMETRIC TRANSFORMATIONS Mehods for geomeric ransformaions and objec modeling in hree dimensions

More information

Recovering Joint and Individual Components in Facial Data

Recovering Joint and Individual Components in Facial Data JOURNAL OF L A E X CLASS FILES, VOL. 14, NO. 8, AUGUS 2015 1 Recovering Join and Individual Componens in Facial Daa Chrisos Sagonas, Evangelos Ververas, Yannis Panagakis, and Sefanos Zafeiriou, Member,

More information

We are IntechOpen, the world s leading publisher of Open Access books Built by scientists, for scientists. International authors and editors

We are IntechOpen, the world s leading publisher of Open Access books Built by scientists, for scientists. International authors and editors We are InechOpen, he world s leading publisher of Open Access books Buil by scieniss, for scieniss 4,000 116,000 120M Open access books available Inernaional auhors and ediors Downloads Our auhors are

More information

An Improved Square-Root Nyquist Shaping Filter

An Improved Square-Root Nyquist Shaping Filter An Improved Square-Roo Nyquis Shaping Filer fred harris San Diego Sae Universiy fred.harris@sdsu.edu Sridhar Seshagiri San Diego Sae Universiy Seshigar.@engineering.sdsu.edu Chris Dick Xilinx Corp. chris.dick@xilinx.com

More information

Moving Object Detection Using MRF Model and Entropy based Adaptive Thresholding

Moving Object Detection Using MRF Model and Entropy based Adaptive Thresholding Moving Objec Deecion Using MRF Model and Enropy based Adapive Thresholding Badri Narayan Subudhi, Pradipa Kumar Nanda and Ashish Ghosh Machine Inelligence Uni, Indian Saisical Insiue, Kolkaa, 700108, India,

More information

Viewpoint Invariant 3D Landmark Model Inference from Monocular 2D Images Using Higher-Order Priors

Viewpoint Invariant 3D Landmark Model Inference from Monocular 2D Images Using Higher-Order Priors Viewpoin Invarian 3D Landmark Model Inference from Monocular 2D Images Using Higher-Order Priors Chaohui Wang 1,2, Yun Zeng 3, Loic Simon 1, Ioannis Kakadiaris 4, Dimiris Samaras 3, Nikos Paragios 1,2

More information

A Fast Stereo-Based Multi-Person Tracking using an Approximated Likelihood Map for Overlapping Silhouette Templates

A Fast Stereo-Based Multi-Person Tracking using an Approximated Likelihood Map for Overlapping Silhouette Templates A Fas Sereo-Based Muli-Person Tracking using an Approximaed Likelihood Map for Overlapping Silhouee Templaes Junji Saake Jun Miura Deparmen of Compuer Science and Engineering Toyohashi Universiy of Technology

More information

Tracking Deforming Objects Using Particle Filtering for Geometric Active Contours

Tracking Deforming Objects Using Particle Filtering for Geometric Active Contours 1470 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 29, NO. 8, AUGUST 2007 Tracking Deforming Objecs Using Paricle Filering for Geomeric Acive Conours Yogesh Rahi, Member, IEEE, NamraaVaswani,

More information

Gender Classification of Faces Using Adaboost*

Gender Classification of Faces Using Adaboost* Gender Classificaion of Faces Using Adaboos* Rodrigo Verschae 1,2,3, Javier Ruiz-del-Solar 1,2, and Mauricio Correa 1,2 1 Deparmen of Elecrical Engineering, Universidad de Chile 2 Cener for Web Research,

More information

Computer representations of piecewise

Computer representations of piecewise Edior: Gabriel Taubin Inroducion o Geomeric Processing hrough Opimizaion Gabriel Taubin Brown Universiy Compuer represenaions o piecewise smooh suraces have become vial echnologies in areas ranging rom

More information

Analysis of Various Types of Bugs in the Object Oriented Java Script Language Coding

Analysis of Various Types of Bugs in the Object Oriented Java Script Language Coding Indian Journal of Science and Technology, Vol 8(21), DOI: 10.17485/ijs/2015/v8i21/69958, Sepember 2015 ISSN (Prin) : 0974-6846 ISSN (Online) : 0974-5645 Analysis of Various Types of Bugs in he Objec Oriened

More information

IntentSearch:Capturing User Intention for One-Click Internet Image Search

IntentSearch:Capturing User Intention for One-Click Internet Image Search JOURNAL OF L A T E X CLASS FILES, VOL. 6, NO. 1, JANUARY 2010 1 InenSearch:Capuring User Inenion for One-Click Inerne Image Search Xiaoou Tang, Fellow, IEEE, Ke Liu, Jingyu Cui, Suden Member, IEEE, Fang

More information

Shortest Path Algorithms. Lecture I: Shortest Path Algorithms. Example. Graphs and Matrices. Setting: Dr Kieran T. Herley.

Shortest Path Algorithms. Lecture I: Shortest Path Algorithms. Example. Graphs and Matrices. Setting: Dr Kieran T. Herley. Shores Pah Algorihms Background Seing: Lecure I: Shores Pah Algorihms Dr Kieran T. Herle Deparmen of Compuer Science Universi College Cork Ocober 201 direced graph, real edge weighs Le he lengh of a pah

More information

Real Time Integral-Based Structural Health Monitoring

Real Time Integral-Based Structural Health Monitoring Real Time Inegral-Based Srucural Healh Monioring The nd Inernaional Conference on Sensing Technology ICST 7 J. G. Chase, I. Singh-Leve, C. E. Hann, X. Chen Deparmen of Mechanical Engineering, Universiy

More information

STRING DESCRIPTIONS OF DATA FOR DISPLAY*

STRING DESCRIPTIONS OF DATA FOR DISPLAY* SLAC-PUB-383 January 1968 STRING DESCRIPTIONS OF DATA FOR DISPLAY* J. E. George and W. F. Miller Compuer Science Deparmen and Sanford Linear Acceleraor Cener Sanford Universiy Sanford, California Absrac

More information

Simultaneous Localization and Mapping with Stereo Vision

Simultaneous Localization and Mapping with Stereo Vision Simulaneous Localizaion and Mapping wih Sereo Vision Mahew N. Dailey Compuer Science and Informaion Managemen Asian Insiue of Technology Pahumhani, Thailand Email: mdailey@ai.ac.h Manukid Parnichkun Mecharonics

More information

MATH Differential Equations September 15, 2008 Project 1, Fall 2008 Due: September 24, 2008

MATH Differential Equations September 15, 2008 Project 1, Fall 2008 Due: September 24, 2008 MATH 5 - Differenial Equaions Sepember 15, 8 Projec 1, Fall 8 Due: Sepember 4, 8 Lab 1.3 - Logisics Populaion Models wih Harvesing For his projec we consider lab 1.3 of Differenial Equaions pages 146 o

More information

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA Audio Engineering Sociey Convenion Paper Presened a he 119h Convenion 2005 Ocober 7 10 New Yor, New Yor USA This convenion paper has been reproduced from he auhor's advance manuscrip, wihou ediing, correcions,

More information

MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES

MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES B. MARCOTEGUI and F. MEYER Ecole des Mines de Paris, Cenre de Morphologie Mahémaique, 35, rue Sain-Honoré, F 77305 Fonainebleau Cedex, France Absrac. In image

More information

Detection and segmentation of moving objects in highly dynamic scenes

Detection and segmentation of moving objects in highly dynamic scenes Deecion and segmenaion of moving objecs in highly dynamic scenes Aurélie Bugeau Parick Pérez INRIA, Cenre Rennes - Breagne Alanique Universié de Rennes, Campus de Beaulieu, 35 042 Rennes Cedex, France

More information

Robot localization under perceptual aliasing conditions based on laser reflectivity using particle filter

Robot localization under perceptual aliasing conditions based on laser reflectivity using particle filter Robo localizaion under percepual aliasing condiions based on laser refleciviy using paricle filer DongXiang Zhang, Ryo Kurazume, Yumi Iwashia, Tsuomu Hasegawa Absrac Global localizaion, which deermines

More information

In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magnetic Field Maps

In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magnetic Field Maps In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magneic Field Maps A. D. Hahn 1, A. S. Nencka 1 and D. B. Rowe 2,1 1 Medical College of Wisconsin, Milwaukee, WI, Unied

More information

Real-Time Avatar Animation Steered by Live Body Motion

Real-Time Avatar Animation Steered by Live Body Motion Real-Time Avaar Animaion Seered by Live Body Moion Oliver Schreer, Ralf Tanger, Peer Eiser, Peer Kauff, Bernhard Kaspar, and Roman Engler 3 Fraunhofer Insiue for Telecommunicaions/Heinrich-Herz-Insiu,

More information

DAGM 2011 Tutorial on Convex Optimization for Computer Vision

DAGM 2011 Tutorial on Convex Optimization for Computer Vision DAGM 2011 Tuorial on Convex Opimizaion for Compuer Vision Par 3: Convex Soluions for Sereo and Opical Flow Daniel Cremers Compuer Vision Group Technical Universiy of Munich Graz Universiy of Technology

More information

Evaluation and Improvement of Region-based Motion Segmentation

Evaluation and Improvement of Region-based Motion Segmentation Evaluaion and Improvemen of Region-based Moion Segmenaion Mark Ross Universiy Koblenz-Landau, Insiue of Compuaional Visualisics, Universiässraße 1, 56070 Koblenz, Germany Email: ross@uni-koblenz.de Absrac

More information

Video Content Description Using Fuzzy Spatio-Temporal Relations

Video Content Description Using Fuzzy Spatio-Temporal Relations Proceedings of he 4s Hawaii Inernaional Conference on Sysem Sciences - 008 Video Conen Descripion Using Fuzzy Spaio-Temporal Relaions rchana M. Rajurkar *, R.C. Joshi and Sananu Chaudhary 3 Dep of Compuer

More information

An Iterative Scheme for Motion-Based Scene Segmentation

An Iterative Scheme for Motion-Based Scene Segmentation An Ieraive Scheme for Moion-Based Scene Segmenaion Alexander Bachmann and Hildegard Kuehne Deparmen for Measuremen and Conrol Insiue for Anhropomaics Universiy of Karlsruhe (H), 76 131 Karlsruhe, Germany

More information

Wheelchair-user Detection Combined with Parts-based Tracking

Wheelchair-user Detection Combined with Parts-based Tracking Wheelchair-user Deecion Combined wih Pars-based Tracking Ukyo Tanikawa 1, Yasuomo Kawanishi 1, Daisuke Deguchi 2,IchiroIde 1, Hiroshi Murase 1 and Ryo Kawai 3 1 Graduae School of Informaion Science, Nagoya

More information

AUTOMATIC 3D FACE REGISTRATION WITHOUT INITIALIZATION

AUTOMATIC 3D FACE REGISTRATION WITHOUT INITIALIZATION Chaper 3 AUTOMATIC 3D FACE REGISTRATION WITHOUT INITIALIZATION A. Koschan, V. R. Ayyagari, F. Boughorbel, and M. A. Abidi Imaging, Roboics, and Inelligen Sysems Laboraory, The Universiy of Tennessee, 334

More information

Projection & Interaction

Projection & Interaction Projecion & Ineracion Algebra of projecion Canonical viewing volume rackball inerface ransform Hierarchies Preview of Assignmen #2 Lecure 8 Comp 236 Spring 25 Projecions Our lives are grealy simplified

More information

IAJIT First Online Publication

IAJIT First Online Publication An Improved Feaure Exracion and Combinaion of Muliple Classifiers for Query-by- ming Naha Phiwma and Parinya Sanguansa 2 Deparmen of Compuer Science, Suan Dusi Rajabha Universiy, Thailand 2 Faculy of Engineering

More information

PART 1 REFERENCE INFORMATION CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONITOR

PART 1 REFERENCE INFORMATION CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONITOR . ~ PART 1 c 0 \,).,,.,, REFERENCE NFORMATON CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONTOR n CONTROL DATA 6400 Compuer Sysems, sysem funcions are normally handled by he Monior locaed in a Peripheral

More information

Lecture 18: Mix net Voting Systems

Lecture 18: Mix net Voting Systems 6.897: Advanced Topics in Crypography Apr 9, 2004 Lecure 18: Mix ne Voing Sysems Scribed by: Yael Tauman Kalai 1 Inroducion In he previous lecure, we defined he noion of an elecronic voing sysem, and specified

More information

Dynamic Depth Recovery from Multiple Synchronized Video Streams 1

Dynamic Depth Recovery from Multiple Synchronized Video Streams 1 Dynamic Deph Recoery from Muliple ynchronized Video reams Hai ao, Harpree. awhney, and Rakesh Kumar Deparmen of Compuer Engineering arnoff Corporaion Uniersiy of California a ana Cruz Washingon Road ana

More information

Optimal Crane Scheduling

Optimal Crane Scheduling Opimal Crane Scheduling Samid Hoda, John Hooker Laife Genc Kaya, Ben Peerson Carnegie Mellon Universiy Iiro Harjunkoski ABB Corporae Research EWO - 13 November 2007 1/16 Problem Track-mouned cranes move

More information

Multi-Target Detection and Tracking from a Single Camera in Unmanned Aerial Vehicles (UAVs)

Multi-Target Detection and Tracking from a Single Camera in Unmanned Aerial Vehicles (UAVs) 2016 IEEE/RSJ Inernaional Conference on Inelligen Robos and Sysems (IROS) Daejeon Convenion Cener Ocober 9-14, 2016, Daejeon, Korea Muli-Targe Deecion and Tracking from a Single Camera in Unmanned Aerial

More information

Quantitative macro models feature an infinite number of periods A more realistic (?) view of time

Quantitative macro models feature an infinite number of periods A more realistic (?) view of time INFINIE-HORIZON CONSUMPION-SAVINGS MODEL SEPEMBER, Inroducion BASICS Quaniaive macro models feaure an infinie number of periods A more realisic (?) view of ime Infinie number of periods A meaphor for many

More information

LAMP: 3D Layered, Adaptive-resolution and Multiperspective Panorama - a New Scene Representation

LAMP: 3D Layered, Adaptive-resolution and Multiperspective Panorama - a New Scene Representation Submission o Special Issue of CVIU on Model-based and Image-based 3D Scene Represenaion for Ineracive Visualizaion LAMP: 3D Layered, Adapive-resoluion and Muliperspecive Panorama - a New Scene Represenaion

More information

4 Error Control. 4.1 Issues with Reliable Protocols

4 Error Control. 4.1 Issues with Reliable Protocols 4 Error Conrol Jus abou all communicaion sysems aemp o ensure ha he daa ges o he oher end of he link wihou errors. Since i s impossible o build an error-free physical layer (alhough some shor links can

More information

Assignment 2. Due Monday Feb. 12, 10:00pm.

Assignment 2. Due Monday Feb. 12, 10:00pm. Faculy of rs and Science Universiy of Torono CSC 358 - Inroducion o Compuer Neworks, Winer 218, LEC11 ssignmen 2 Due Monday Feb. 12, 1:pm. 1 Quesion 1 (2 Poins): Go-ack n RQ In his quesion, we review how

More information

Real-Time Non-Rigid Multi-Frame Depth Video Super-Resolution

Real-Time Non-Rigid Multi-Frame Depth Video Super-Resolution Real-Time Non-Rigid Muli-Frame Deph Video Super-Resoluion Kassem Al Ismaeil 1, Djamila Aouada 1, Thomas Solignac 2, Bruno Mirbach 2, Björn Oersen 1 1 Inerdisciplinary Cenre for Securiy, Reliabiliy, and

More information

Time Expression Recognition Using a Constituent-based Tagging Scheme

Time Expression Recognition Using a Constituent-based Tagging Scheme Track: Web Conen Analysis, Semanics and Knowledge Time Expression Recogniion Using a Consiuen-based Tagging Scheme Xiaoshi Zhong and Erik Cambria School of Compuer Science and Engineering Nanyang Technological

More information

Definition and examples of time series

Definition and examples of time series Definiion and examples of ime series A ime series is a sequence of daa poins being recorded a specific imes. Formally, le,,p be a probabiliy space, and T an index se. A real valued sochasic process is

More information

Landmarks: A New Model for Similarity-Based Pattern Querying in Time Series Databases

Landmarks: A New Model for Similarity-Based Pattern Querying in Time Series Databases Lmarks: A New Model for Similariy-Based Paern Querying in Time Series Daabases Chang-Shing Perng Haixun Wang Sylvia R. Zhang D. So Parker perng@cs.ucla.edu hxwang@cs.ucla.edu Sylvia Zhang@cle.com so@cs.ucla.edu

More information

AML710 CAD LECTURE 11 SPACE CURVES. Space Curves Intrinsic properties Synthetic curves

AML710 CAD LECTURE 11 SPACE CURVES. Space Curves Intrinsic properties Synthetic curves AML7 CAD LECTURE Space Curves Inrinsic properies Synheic curves A curve which may pass hrough any region of hreedimensional space, as conrased o a plane curve which mus lie on a single plane. Space curves

More information

Track-based and object-based occlusion for people tracking refinement in indoor surveillance

Track-based and object-based occlusion for people tracking refinement in indoor surveillance Trac-based and objec-based occlusion for people racing refinemen in indoor surveillance R. Cucchiara, C. Grana, G. Tardini Diparimeno di Ingegneria Informaica - Universiy of Modena and Reggio Emilia Via

More information

A time-space consistency solution for hardware-in-the-loop simulation system

A time-space consistency solution for hardware-in-the-loop simulation system Inernaional Conference on Advanced Elecronic Science and Technology (AEST 206) A ime-space consisency soluion for hardware-in-he-loop simulaion sysem Zexin Jiang a Elecric Power Research Insiue of Guangdong

More information

MARSS Reference Sheet

MARSS Reference Sheet MARSS Reference Shee The defaul MARSS model (form="marxss") is wrien as follows: x = B x 1 + u + C c + w where w MVN( Q ) y = Z x + a + D d + v where v MVN( R ) x 1 MVN(π Λ) or x MVN(π Λ) c and d are inpus

More information

MOTION TRACKING is a fundamental capability that

MOTION TRACKING is a fundamental capability that TECHNICAL REPORT CRES-05-008, CENTER FOR ROBOTICS AND EMBEDDED SYSTEMS, UNIVERSITY OF SOUTHERN CALIFORNIA 1 Real-ime Moion Tracking from a Mobile Robo Boyoon Jung, Suden Member, IEEE, Gaurav S. Sukhame,

More information

Learning nonlinear appearance manifolds for robot localization

Learning nonlinear appearance manifolds for robot localization Learning nonlinear appearance manifolds for robo localizaion Jihun Hamm, Yuanqing Lin, and Daniel. D. Lee GRAS Lab, Deparmen of Elecrical and Sysems Engineering Universiy of ennsylvania, hiladelphia, A

More information

THE goal of this work is to develop statistical models for

THE goal of this work is to develop statistical models for IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 32, NO. 4, APRIL 2010 579 Nonsaionary Shape Aciviies: Dynamic Models for Landmark Shape Change and Applicaions Samarji Das, Suden Member,

More information

CONTEXT MODELS FOR CRF-BASED CLASSIFICATION OF MULTITEMPORAL REMOTE SENSING DATA

CONTEXT MODELS FOR CRF-BASED CLASSIFICATION OF MULTITEMPORAL REMOTE SENSING DATA ISPRS Annals of he Phoogrammery, Remoe Sensing and Spaial Informaion Sciences, Volume I-7, 2012 XXII ISPRS Congress, 25 Augus 01 Sepember 2012, Melbourne, Ausralia CONTEXT MODELS FOR CRF-BASED CLASSIFICATION

More information

Precise Voronoi Cell Extraction of Free-form Rational Planar Closed Curves

Precise Voronoi Cell Extraction of Free-form Rational Planar Closed Curves Precise Voronoi Cell Exracion of Free-form Raional Planar Closed Curves Iddo Hanniel, Ramanahan Muhuganapahy, Gershon Elber Deparmen of Compuer Science Technion, Israel Insiue of Technology Haifa 32000,

More information

Proceeding of the 6 th International Symposium on Artificial Intelligence and Robotics & Automation in Space: i-sairas 2001, Canadian Space Agency,

Proceeding of the 6 th International Symposium on Artificial Intelligence and Robotics & Automation in Space: i-sairas 2001, Canadian Space Agency, Proceeding of he 6 h Inernaional Symposium on Arificial Inelligence and Roboics & Auomaion in Space: i-sairas 00, Canadian Space Agency, S-Huber, Quebec, Canada, June 8-, 00. Muli-resoluion Mapping Using

More information

Motion Level-of-Detail: A Simplification Method on Crowd Scene

Motion Level-of-Detail: A Simplification Method on Crowd Scene Moion Level-of-Deail: A Simplificaion Mehod on Crowd Scene Absrac Junghyun Ahn VR lab, EECS, KAIST ChocChoggi@vr.kais.ac.kr hp://vr.kais.ac.kr/~zhaoyue Recen echnological improvemen in characer animaion

More information

The Impact of Product Development on the Lifecycle of Defects

The Impact of Product Development on the Lifecycle of Defects The Impac of Produc Developmen on he Lifecycle of Rudolf Ramler Sofware Compeence Cener Hagenberg Sofware Park 21 A-4232 Hagenberg, Ausria +43 7236 3343 872 rudolf.ramler@scch.a ABSTRACT This paper invesigaes

More information

Deep Appearance Models for Face Rendering

Deep Appearance Models for Face Rendering Deep Appearance Models for Face Rendering STEPHEN LOMBARDI, Facebook Realiy Labs JASON SARAGIH, Facebook Realiy Labs TOMAS SIMON, Facebook Realiy Labs YASER SHEIKH, Facebook Realiy Labs Deep Appearance

More information

Detection Tracking and Recognition of Human Poses for a Real Time Spatial Game

Detection Tracking and Recognition of Human Poses for a Real Time Spatial Game Deecion Tracking and Recogniion of Human Poses for a Real Time Spaial Game Feifei Huo, Emile A. Hendriks, A.H.J. Oomes Delf Universiy of Technology The Neherlands f.huo@udelf.nl Pascal van Beek, Remco

More information

TrackNet: Simultaneous Detection and Tracking of Multiple Objects

TrackNet: Simultaneous Detection and Tracking of Multiple Objects TrackNe: Simulaneous Deecion and Tracking of Muliple Objecs Chenge Li New York Universiy cl2840@nyu.edu Gregory Dobler New York Universiy greg.dobler@nyu.edu Yilin Song New York Universiy ys1297@nyu.edu

More information