Margin-Constrained Multiple Kernel Learning Based Multi-Modal Fusion for Affect Recognition

Size: px
Start display at page:

Download "Margin-Constrained Multiple Kernel Learning Based Multi-Modal Fusion for Affect Recognition"

Transcription

1 Margn-Constraned Multple Kernel Learnng Based Mult-Modal Fuson for Affect Recognton Shzh Chen and Yngl Tan Electrcal Engneerng epartment The Cty College of New Yor New Yor, NY USA {schen, Abstract Recent advances n multple-ernel learnng (MKL) show the effectveness to fuse multple base features n obect detecton and recognton. However, MKL tends to select only the most dscrmnatve base features but gnore other less dscrmnatve base features whch may provde complementary nformaton. Moreover, MKL usually employ Gaussan RBF ernels to transform each base feature to ts hgh dmensonal space. Generally, base features from dfferent modaltes requre dfferent ernel parameters for obtanng the optmal performance. Therefore, MKL may fal to utlze the maxmum dscrmnatve power of all base features from multple modaltes at the same tme. In order to address these ssues, we propose a margn-constraned multple-ernel learnng (MCMKL) method by extendng MKL wth margn constrants and applyng dmensonally normalzed RBF (NRBF) ernels for applcaton of mult-modal feature fuson. The proposed MCMKL method learns weghts of dfferent base features accordng to ther dscrmnatve power. Unle the conventonal MKL, MCMKL ncorporates less dscrmnatve base features by assgnng smaller weghts when constructng the optmal combned ernel, so that we can fully tae the advantages of the complementary features from dfferent modaltes. We valdate the proposed MCMKL method for affect recognton from face and body gesture modaltes on the FABO dataset. Our extensve experments demonstrate favorable results as compared to the exstng wor, and MKL-based approach. Keywords-multmodal fuson; affect recognton; multple ernel learnng; I. INTROUCTION Recent research demonstrate that affect recognton from multple modaltes acheves better performance [6, 0,, 5, ]. However, most lteratures n affect recognton are ether only based on features extracted from sngle modalty [, 6, ] or ust usng smply concatenated features from dfferent modaltes [6, 0, 5, ]. How to effectvely fuse features from dfferent modaltes s stll an open queston. The most popular methodology at feature level fuson s the smple concatenaton of feature vectors from dfferent modaltes to form a large feature vector [6, 0, 5, ]. However, ths fuson method requres a careful desgn for selectons of features and parameters, such as feature dmenson etc. Ths s essentally the manual feature selecton. Shan et al. [5] apply the Canoncal Correlaton Analyss (CCA) to proect facal features and body gesture features nto a low dmensonal space whch maxmzes ther correlaton. Then the authors smply concatenate the proected feature vectors together to tran a Support Vector Machne (SVM) classfer for affect recognton. However, t s dffcult to extend ths method to more than two types of features, or base features hereafter, snce t needs to fnd the correlated space between a par of base features. In ths paper, we defne base features as a set of base descrptors, whch can be combned to form an optmal ernel. In practce, t s very lely to have more than two base features for affect recognton due to the problem complexty. Gunes and Pccard [0] select frames, whch are the common apex frames from both face and body gesture modaltes for affect recognton, and then perform a drect concatenaton to combne base features from both modaltes. However, the apex frame selecton s based on the nowledge of temporal dynamcs, whch s usually very dffcult to predct n advance. The drect concatenaton fuson method s vulnerable to the contamnaton of less dscrmnatve base features, especally those wth large feature dmensons. The multple-ernel learnng (MKL) [, 3, 8] s able to partally elmnate some drawbacs of the drect concatenaton fuson method. MKL provdes sheldng from the contamnaton of the less dscrmnatve base features by assgnng very large weghts to the most dscrmnatve base features. It has recently shown the effectveness to fuse multple base features n obect detecton and recognton [8, 9]. However, MKL tends to select only the most dscrmnatve base features and gnore other less dscrmnatve base features. Therefore, MKL method cannot fully tae the advantages of all types of base features from multple modaltes, whch provde complementary nformaton. Moreover, MKL usually employs Gaussan RBF ernels for mappng each base feature to ts hgh dmensonal space H. Generally, base features from dfferent modaltes requre dfferent ernel parameters to acheve ther optmal performance. One of the reasons s due to the sgnfcant dfferent feature dmensons from multple modaltes. Therefore, MKL may not utlze the maxmum dscrmnatve power of all types of features from multple modaltes at the same tme.

2 In order to address these ssues, we propose a margnconstraned multple-ernel learnng (MCMKL) by applyng () addtonal margn constrants, and () dmensonally normalzed RBF ernels (NRBF). The margn of a separatng hyper-plane n SVM lterature [3] s defned as the perpendcular dstance between the support vectors of two classes. The large margn, obtaned n the tranng set, usually ndcates that the underlyng feature s dscrmnatve. If the underlyng feature s not dscrmnatve, the margn s usually small. So we use the margn to measure the dscrmnatve power of each base feature. These margns then serve as rough gude to MKL when learnng each base feature s weght. By dmensonally normalzng RBF ernels, MCMKL elmnates the nfluence of the feature dmenson dfference n multple modaltes. Then, the optmal hgh dmensonal space H of each base feature from dfferent modaltes corresponds to smlar ernel parameter. Therefore, MCMKL s able to utlze the maxmum dscrmnatve power of all feature types from multple modaltes. Unle conventonal MKL method, our proposed MCMKL s able to learn the most dscrmnatve base features whle stll consderng other base features, whch are less dscrmnatve, but can potentally provde complementary nformaton. We apply MCMKL method on affect recognton from multple modaltes (e.g. both face and body gesture). The extensve expermental results on the FABO (Face and Body Gesture) facal expresson database [9] demonstrate the effectveness of the proposed method for mult-modal feature fuson. II. MARGIN-CONSTRAINE MULTIPLE KERNEL LEARNING A. Multple Kernel Learnng Gven a set of base features and ther assocated base ernels K, we want to fnd the optmal ernel combnaton K opt d * K, where d s the weght for the th base feature. The ernel combnaton K opt approxmates the best trade-off between the dscrmnatve power and the nvarance for a specfc applcaton. Equatons () to (3) show the obectve cost functon f and ts constrants for the multple-ernel learnng (MKL) proposed n [8]. Mn f w w, ξ, d subect to + C ξ + σ y ( w Φ( x ) + b) + ξ 0 () ξ 0 ; d 0 ; d () Ad p (3) where Φ( x ) s the combned features correspondng to K opt n a hgh dmensonal space for sample x, whch s shown n Eq. (4). Equvalently, K opt can be expressed n Eq. (5), where Φ x ) Φ ( x ) forms the th base ernel K. ( K opt ( x, x ) Φ( x ) Φ( x ) (4) K ( x, x ) d * Φ ( x ) Φ ( x ) (5) opt The optmzaton can be carred out n a SVM framewor subect to addtonal regularzaton term of weght d n the obectve functon. In order to handle large scale problems nvolvng many base ernels, the mnmax optmzaton strategy [5, 3, 8] s used n two teraton steps. In the frst step, feature weght d s fxed,.e., K opt d * K s fxed. Then, the optmzaton problem of Eq. () can be solved by any standard SVM solver usng ts dual form as n Eq. (6) snce the term σ d s smply a constant. Max f subect to y y ( x, x ) + σ, K d (6) 0 C ; y 0 (7) In the second teraton step wth the fxed α, proected gradent descent s employed to fnd updated feature weghts d as shown n Eqs. (8) and (9). f d d new f d opt σ y y K ( x, x ) (8), old f d (9) d These two teraton steps are repeated untl converge or the maxmum number of teratons s reached. The fnal weghts of base features can be determned. Then we tran SVM classfers usng the optmal combned ernel accordng the fnal weghts of base features. The class label of a new sample x may be determned by the sgn of Eq. (0). g( x) y d K ( s, x) + b (0) where s s the support vector. The mult-class problem can be solved by one vs. one or one vs. all strategy smlar to SVM. MKL method provdes an elegant framewor to fuse many base features by assgnng larger feature weght to the most dscrmnatve base feature. Compared to the drect concatenaton method, MKL can avod the contamnaton from less dscrmnatve base features, especally when those features have large dmensons. However, MKL method tends to select very few base features from the feature pool. It often only selects one or two base features, whch are dscrmnatve at a partcular hgh dmensonal space H. Moreover, the ernel parameter

3 assocated wth the optmal hgh dmensonal space H may be sgnfcantly dfferent for the base features from dfferent modaltes. Therefore, tradtonal MKL cannot utlze the maxmum dscrmnatve power of each base feature at the same tme. B. Margn Constrants To address these ssues, we propose a Margn-Constraned Multple Kernel Learnng (MCMKL) method. Ths s motvated by the observatons that base feature whch s more dscrmnatve usually fnds a hyper-plane wth larger margn to separate support vectors of opposte classes durng tranng of SVM machnes. A hyper-plane of base feature a n Fgure (a) has a larger margn than that of base feature b n Fgure (b) to separate the class of sold dot from the class of trangle. Ths suggests that the base feature a s more dscrmnatve than the base feature b for the classfcaton of the sold dot and the trangle class. (a) (b) Fgure : (a) The hyper-plane of base feature a has a large separaton margn to separate sold dot class and trangle class; (b) The hyper-plane of base feature b has a small margn to separate sold dot class and trangle class. Therefore, the separaton margn for each base feature n ts hgh dmensonal space provdes a rough measurement on the base feature s dscrmnatve power. Nevertheless, these rough measurements can effectvely gude MKL when searchng for the optmal feature combnaton. The separaton margn for each base feature can be calculated usng Eq. () as the nversed square root of ts own obectve cost functon. m w f w + C ξ +σ d () After obtanng the separaton margn m for each base feature, we select one of the base feature as the reference base feature, whch has the feature weght of d s and the margn m s. The weght d of th base feature s constraned n the range, whch has the lower bound of LB and the upper bound of UB accordng to the margn rato between m s and m durng tranng. LB and UB can be calculated as n Eq. (3). The addtonal weght constrants n Eq. () are enforced durng the multple-ernel learnng. LB d UB () n m m LB * d s m ; UB * ds *( + δ ) (3) s ms n where n s a parameter that controls the margn senstvty on the feature weght rato between d and d s. As n ncreases, the values of LB and UB become more senstve to the rato of m and m s. δ s a constant to control the range wdth of the feature weght d. In our experments, we set n to.5 and δ to. C. mensonally Normalzed Kernel Gaussan RBF ernel s one of the most popular non-lnear ernels due to ts excellent performance n numerous applcatons. It s defned n Eq. (4). K ( x, x ) exp( γ ( x x ) (4) q, q, q ) where x and x are the th sample and the th sample along wth x,q and x,q as the q th element n a feature vector. s the sample s feature dmenson. γ s the RBF ernel parameter, whch determnes the mappng from a low dmensonal feature space L to a hgh dmensonal space H. Assumng that feature vectors x and x have been properly normalzed between 0 and along each feature dmenson [4], the ernel value decreases when the feature dmenson ncreases at a fxed γ as shown n Eq. (4). Hence, Eq. (4) suggests the nverse relatonshp between the optmal γ and the feature dmenson. Ths ntuton s confrmed n our experments, whch wll be analyzed n Secton 4. In MKL fuson, base features from dfferent modaltes may have sgnfcantly dfferent feature dmensons, whch wll result very dfferent optmal γ values for each base feature. Therefore, MKL cannot utlze the maxmum dscrmnatve power of all base features from dfferent modaltes at the same tme. We can treat γ as a feature selecton parameter n MKL, whch select only few base features at a tme. Ths ntuton also explans the observatons reported n [8] that MKL tends to select only very few most dscrmnatve base features. Therefore, MKL cannot tae the full advantages of all types of features from multple modaltes. Based on these observatons, we propose a dmensonally normalzed RBF ernel (NRBF) whch s defned n Eq. (5). γ K ( x, x ) exp( ( x, q x, q ) ) (5) q Ths normalzaton step s essental to elmnate the effect of feature dmenson on γ selecton, so that all base features have a smlar optmal γ. Therefore, MCMKL can utlze the maxmum dscrmnatve power of all base features from multple modaltes. III. MULTI-MOAL FUSION FOR AFFECT RECOGNITION Affect recognton from multple modaltes s a challengng problem. Our study focuses on fuson of features from vsual modaltes,.e., face and body gesture modalty.

4 fferent from conventonal approaches to fuse features from multple modaltes, whch smply concatenate all feature vectors from dfferent sources together and feed the concatenated feature vector nto a classfer, such as SVM, we apply a margn-constraned multple-ernel learnng (MCMKL) method to fuse features from both face and body gesture modaltes. MCMKL can effectvely combne all types of features for affect recognton by assgnng an approprate feature weght to each type of features and calculate the optmal ernel for affect recognton. A. Overvew of MCMKL-based Affect Recognton Fgure shows an overvew of our affect recognton system, whch conssts of fve maor parts,.e., facal feature extracton, body gesture feature extracton, expresson temporal segmentaton, temporal normalzaton, and MCMKL-based classfcaton. Two types of facal features,.e., Image-HOG and MHI- HOG [6] are extracted n our experments. Here, HOG stands for Hstogram of Gradents [8], and MHI stands for Moton Hstory Image [, 7]. Image-HOG features capture facal appearance changes, whle MHI-HOG features represent facal moton nformaton. Four types of gesture features are extracted, whch nclude locaton features, moton area features, Image-HOG features, and MHI-HOG features around both hands. Each expresson n vdeo sequences can be frst temporally segmented nto onset, apex, offset and neutral phases [6]. Then, we perform a temporal normalzaton procedure to handle dfferent temporal resolutons of expressons. Fnally MCMKL method s employed to fnd the optmal feature combnaton and recognze affects. B. Facal Features Actve Shape Model [7, 0] s frst appled to trac 53 facal landmar ponts ncludng brows, eyes, nose, mouth, and face contour, as shown n Fgure 3(a). Then we locate the correspondng postons of the facal ponts n the Moton Hstory Image (MHI), as shown n Fgure 3(b). The next step s to extract Image-HOG and MHI-HOG features on orgnal vdeo frames and the correspondng MHI mages respectvely. We use 48 by 48 pxels patches wth the number of orentaton bn equals to 6 and 8 for the Image-HOG and the MHI-HOG features respectvely. The MHI mage captures moton nformaton of each selected facal pont, whle the orgnal vdeo frame conveys the appearance nformaton. Fnally, we concatenate the Image-HOG descrptor of all the 53 facal ponts and apply Prncpal Component Analyss (PCA) to reduce the feature dmenson of the concatenated Image-HOG feature from 86 to 40. Smlarly, we can obtan the MHI- HOG descrptor for the correspondng frame and reduce the feature dmenson of the concatenated MHI-HOG from 386 down to 40 for each frame. (a) (b) Fgure 3: (a) Facal landmar ponts tracng; (b) Moton Hstory Image. C. Body Gesture Features To extract body gesture features, we frst trac both hands and head n an expresson vdeo. The head poston s smply the center pont of the facal ponts from the ASM model (see Fgure 4(b)). To trac hands, we apply a sn color detecton [] followed by the removal of the face regons, whch has already been traced by the ASM model as shown n Fgure 4(a) and 4(b). (a) (b) (c) Fgure 4: (a) sn color detecton; (b) head and hand poston n the orgnal vdeo frame; (c) head and hand postons n the MHI mage. In addton to the postons of head and hands, we also calculate the moton areas (e.g. the numbers of moton pxels n MHI mage) wthn the detected regons of head and hands. Fgure 4(c) shows the head and hand regons n a MHI mage. We further extract Image-HOG and MHI-HOG features n hand regons by unformly samplng nterest ponts. Then a bag of words representaton wth the codeboo sze of 80 s used to descrbe the dstrbuton of Image-HOG and MHI-HOG features of hand regons. Fnally, we perform PCA to reduce ther feature dmensons.. Temporal Segmentaton An expresson s a sequence of facal movements whch can be roughly descrbed by neutral, onset, apex and offset temporal segments. Fgure : The overvew of our proposed MCMKL-based multmodal fuson for affect recognton through both face and body gestures. Fgure 5: temporal segmentaton of an expresson vdeo.

5 Fgure 5 shows a sample of the ground truth temporal segmentaton of an expresson vdeo. The temporal segmentaton procedure s necessary to accurately model the expresson dynamcs, whch has been proven crucal for facal behavor nterpretaton [4]. In our experments, we smply use the ground truth temporal segmentaton and the affect recognton s performed on the complete expresson cycle,.e., onset, apex and offset. E. Temporal Normalzaton In general, the temporal resoluton of an expresson s generally dfferent when performed by dfferent people. Even same expresson performed by same person at a dfferent tme, the temporal resoluton may not be the same. In order to resolve ths tme resoluton ssue n expresson vdeos, we adopt the temporal normalzaton approach by normalzng all types of features over a complete expresson cycle. The temporal normalzaton over an expresson cycle can be easly mplemented by lnear nterpolaton over frame s feature vector along the temporal drecton. F. MCMKL Based Mult-Modal Feature Fuson Features from multple modaltes may have dfferent forms. Therefore dmensons of dfferent types of features may vary sgnfcantly. Our proposed margn-constraned multple ernel learnng (MCMKL) method can effectvely fuse all base features from dfferent modaltes,.e., face and body gesture modalty, by assgnng a feature weght to each base feature. We concatenate the Image-HOG and the MHI-HOG of facal ponts as one base feature,.e., the face feature. The other four base features are from the gesture channel,.e., locaton, moton area, and both hands Image-HOG and MHI-HOG features. Usng the margn of each ndvdual base feature as a gude, along wth the NRBF to synchronze the optmal ernel parameter, the MCMKL learns the optmal combned ernel by selectng a proper weght for each base feature durng the fuson. For our mult-classes applcaton, we choose one vs. one classfcaton, and then usng the maxmum votng scheme to label testng samples. number of vdeos used n our experment s 55 and each vdeo has to 4 complete expresson cycles. We randomly dvde the vdeos nto three subsets. Two subsets are used n tranng and the remanng subset s used n testng. No same vdeo appears n both tranng and testng. But same subect may appear n both tranng and testng due to the random selecton process. Fgure 6: sample vdeo n FABO database recorded by body (top) and face (bottom) camera; B. Comparson to Exstng Wor and MKL In order to evaluate the effectveness of the proposed MCMKL fuson method, we compare t to the most recent wor on the FABO database [6] by usng same features, and same tranng and testng dataset. We further compare MCMKL wth MKL method. The performance of the comparson s dsplayed n Fgure 7. The fve base features are used n our experments, whch nclude face feature, locaton feature, moton area feature, Image-HOG and MHI-HOG feature of both hands. The face feature s the concatenaton of the Image-HOG and the MHI- HOG from the face modalty. The Table shows the correspondng feature dmenson for each base feature. These base features are fused through the concatenaton, MKL, and MCMKL methods. Table : The feature dmenson for each base feature,.e., Face, Loc (locaton), MA (moton area), Img-HOG (Image-HOG from gesture), and MHI-HOG (from gesture). Base Face Loc MA Img-HOG MHI-HOG Feature menson IV. EXPERIMENTS A. Expermental Setups We use a b-modal face and body gesture database,.e., FABO database n our experments [9]. The database s collected usng two cameras,.e., one for face and one for body gesture n a laboratory envronment. A sample vdeo s shown n Fgure 6. However, we only employ the vdeos captured by the body camera to extract features for both modaltes, snce the vdeos from the body camera already contan both face and body gesture nformaton. After removng the categores n the database wth very small number of samples, there are 8 expresson categores,.e., Anger, Anxety, Boredom, sgust, Fear, Happness, Puzzlement, and Uncertanty. The total Fgure 7: The average performance of the top rans by sweepng ernel parameter log(γ) from -5 to 8 for each of the three methods,.e., concatenaton (cvpr4hb ), MKL, and MCMKL. To mae a far comparson, we sweep ernel parameter log(γ) from -5 to 8, and select the top performances for each fuson method. Then we ran these performances by a descendng order of ther accuracy. We repeat same experment

6 for three dfferent subsets and the average performances are reported n Fgure 7. Fgure 8 shows more detals of the ran comparson,.e., the comparson of the best performances for the three fuson methods: MCMKL, MKL, and drect feature concatenaton (cvpr4hb ). MCMKL outperforms the other two methods over all the top rans, as shown n Fgure 7. If we loo at the ran comparson n Fgure 8, we can see that the proposed MCMKL acheves better performance than the concatenaton method. Note that the fve base features have been carefully selected, and the parameters, e.g., PCA proecton dmensons etc. are also carefully chosen for the concatenaton fuson method n [6]. On the other hand, MCMKL effectvely select those feature vectors, and t can stll outperform the concatenaton fuson method by the average of.3%. MKL method gnores all the gesture features,.e., locaton, and moton area etc., even though these gesture features have been proven to provde complementary nformaton to the face feature [6]. As shown n Fgure 9, the proposed MCMKL obtans a more reasonable feature weght dstrbuton. Smlar to MKL methods, t recognzes the face feature as the most dscrmnatve base feature by assgnng the largest feature weght of 48%. At the same tme, t also ncorporates other less dscrmnatve gesture features accordng to ther dscrmnatve power. (a) Fgure 8: (a) The best average performance by sweepng ernel parameter log(γ) from -5 to 8 for each fuson method,.e., cvpr4hb, MKL, and MCMKL; (b) The best performance of the three fuson method n three dfferent testng subsets. Our proposed MCMKL outperforms tradtonal MKL method by an average recognton rate of 5.7% on these base features, as shown n Fgure 8(a). Fgure 8(b) shows the performance comparson over three dfferent testng subsets. C. Evaluate Feature Weght strbuton In ths secton we verfy that our proposed MCMKL s more effectve than MKL to ncorporate less dscrmnatve features, whch provde complementary nformaton to the base features wth the maxmum dscrmnatve power. We select the ernel parameter γ of -5 for MKL and - for MCMKL method, whch yeld the best performance for MKL and MCMKL method respectvely. Snce we choose one vs. one strategy for our mult-class expresson classfcaton, the total number of models we need C to tran s n, where n s the total number of expresson classes n the dataset. Therefore, we have traned 8 models for 8 categores of expressons, n whch each model contans one set of feature weghts for the base features,.e., face, locaton (loc), moton area (MA), and both hands Image-HOG (mghog) and MHI-HOG (mhhog) features. Then we tae the mean feature weght of the 8 models over each base feature, followed by the proper normalzaton. Fgure 9 shows the dstrbuton of average feature weghts over the 5 base feature types for MKL and MCMKL method. As expected, MKL selects only the most dscrmnatve base feature,.e., face feature. More specfcally, t assgns more than 98% of the total feature weghts to the face feature. The (b) Fgure 9: Comparng the average feature weght dstrbuton of the 5 base features,.e., face, locaton (loc), moton area (MA), and both hands Image-HOG (mghog) and MHI-HOG (mhhog) features, for MKL and MCMKL methods. Fgure 9 has verfed the effectveness of the proposed margn constrants and the dmensonally normalzed RBF ernel (NRBF). It s obvous that the addtonal constrants on the feature weghts accordng to the separaton margn of each base feature can enforce the model to assgn small weghts to the less dscrmnatve base features. However, t may not be ntutve how the NRBF contrbute to a more reasonable feature weght assgnment. Fgure 0: The effect of feature dmenson of three base features over the selecton of the optmal RBF ernel parameter γ. Before we provde such ntuton, we examne the relatonshp between the optmal γ value and feature dmenson expermentally. We select three base features,.e., facal pont s MHI-HOG, the facal pont s Image-HOG and the locaton feature. Then we manually vary the PCA dmenson of the Image-HOG and the MHI-HOG, or the number of normalzaton frames of the locaton feature, so that ther feature dmensons can be gradually ncreased. Then we use SVM s 5-fold cross valdaton to fnd out the optmal ernel parameter γ for each of the three base features at the selected feature dmenson. Fgure 0 has suggested the nverse relatonshp between the optmal γ value and the feature dmenson, whch has verfed our analyss n secton. In the experments of the last secton, the most dscrmnatve feature,.e., the face feature, has the optmal γ of -5. Snce other less dscrmnatve gesture feature has much

7 smaller feature dmenson as we can see from Table. Ther optmal γ value s much larger. Therefore, at the γ of -5, the other gesture features has almost no dscrmnatve power snce ther optmal γ values are very far away from -5. Therefore, MKL method assgns almost zero feature weghts to other gesture features. After we perform the dmensonally normalzaton as n Eq. (5). The optmal γ values become very close for dfferent base features regardless the dfferences of ther feature dmensons. Therefore, MCMKL can utlze the maxmum dscrmnatve power of all base features at the same tme. That s another reason why MCMKL method can ncorporate other less dscrmnatve base features, whch provde complementary nformaton.. Contamnaton from Less scrmnatve Features In ths secton, we examne the contamnaton from the less dscrmnatve base features, partcularly those wth large feature dmensons. From the feature weght dstrbuton n Fgure 9, we now that the Image-HOG and MHI-HOG of hands are the least dscrmnatve features. So we ntentonally ncrease ther feature dmenson to 00 by ncludng more PCA dmensons. At the same tme, we also decrease the dmenson of the most dscrmnatve feature,.e., the face feature, down to 90. Now, we also sweep ernel parameter log(γ) and select the top 0 performances for each fuson method,.e., concatenaton, MKL, and MCMKL. Then we ran these 0 performances by the descendng order of ther accuracy. The expermental results are shown n Fgure. We observe that the ran result of the MCMKL method outperforms the concatenaton fuson method by almost 0%, whch ndcate that MCMKL method s more effectve to sheld the contamnaton from the less dscrmnatve base features, as compared wth the concatenaton fuson method. V. CONCLUSION In ths paper, we have proposed a margn-constraned multple-ernel learnng (MCMKL) method whch extends the multple-ernel learnng (MKL) method by constranng feature weght range accordng to the separaton margn of each base feature. The dmensonally normalzed RBF ernel (NRBF) s also proposed and employed n MCMKL n order to fuse the features from multple modaltes, whch s possble to have very dfferent feature dmensons. Our expermental results demonstrate favorable results as compared to the stateof-the-art results on the FABO database. We also demonstrate the sgnfcant mprovement as compared to the conventonal MKL method. Fgure : The average performance of the top 0 rans by sweepng ernel parameter γ for each of the three methods,.e., concatenaton (cvpr4hb ), MKL, and MCMKL. ACKNOLEGEMENT Ths wor was supported n part by NSF IIS REFERENCES [] F. Bach, G. Lancret, M. Jordan, Multple ernel learnng, conc dualty and the SMO algorthm, NIPS, 004. [] A. Bobc and J. avs, The recognton of human movement usng temporal templates. PAMI, 00. [3] C. Burges, A tutoral on support vector machnes for pattern recognton, ata Mnng and Knowledge scovery, (), 67, 998. [4] C. Chang and C. Ln, LIBSVM : a lbrary for support vector machnes, 00. Software avalable at [5] O. Chapelle, V. Vapn, O. Bousquet, S. Muheree, Choosng Multple Parameters for Support Vector Machnes, Machne Learnng, 00. [6] S. Chen, Y. Tan, Q. Lu,. Metaxas, Recognzng Expressons from Face and Body Gesture by Temporal Normalzed Moton and Appearance Features, IEEE Int'l Conf. Computer Vson and Pattern Recognton worshop for Human Communcatve Behavor Analyss (CVPR4HB). 0. [7] T. Cootes, C. Taylor,. Cooper and J. Graham, Actve Shape Models Ther Tranng and Applcaton, Computer Vson and Image Understandng, 995. [8] N. alal, B. Trggs, Hstogram of Orented Gradents for Human etecton, CVPR 005 [9] H. Gunes and M. Pccard, A bmodal face and body gesture database for automatc analyss of human nonverbal affectve behavor, Internatonal Conferenece Pattern Recognton, 006. [0] H. Gunes and M. Pccard, Automatc Temporal Segment etecton and Affect Recognton From Face and Body splay, IEEE Transacton on Systems, Man and Cybernetcs Part B: Cybernetcs, Vol. 39, NO [] J. Kovac, P. Peer, F. Solna, Human Sn Colour Clusterng for Face etecton, EUROCON Computer as a Tool, 003. [] M. Pantc, L. Rothrantz, Automatc Analyss of Facal Expressons: The State of the Art, PAMI, 000. [3] A. Raotomamony, F. Bach, S. Canu, Y. Grandvalet, More Effcency n Multple Kernel Learnng, ICML, 007. [4] K. Schmdt and J. Cohn, Human Facal Expressons as Adaptatons: Evolutonary Questons n Facal Expresson Research, Yearboo of Physcal Anthropology, 00. [5] C. Shan, S. Gong and P. McOwan, Beyond facal expressons: learnng human emoton from body gestures, Brtsh Machne Vson Conference, 007. [6] Y. Tan, T. Kanade, J. Cohn, Facal Expresson Analyss, Handboo of Face Recognton, pp , Sprnger, 005. [7] Y. Tan, L. Cao, Z. Lu, and Z. Zhang, Herarchcal Fltered Moton for Acton Recognton n Crowded Vdeos, IEEE Transactons on Systems, Man, and Cybernetcs--Part C: Applcatons and Revews, 0. [8] M. Varma,. Ray, Learnng The scrmnatve Power-Invarance Trade-Off, ICCV, 007. [9] A. Vedald, V. Gulshan, M. Varma, A. Zsserman, Multple Kernels for Obect etecton, ICCV, 009. [0] Y. We, Research on Facal Expresson Recognton and Synthess, Master Thess, 009, software avalable at: [] Z. Zeng, M. Pantc, G. Rosman, T. Huang, A Survey of Affect Recognton Methods: Audo, Vsual, and Spontaneous Expressons, PAMI, 009. [] X. Zhou, B. Bhanu, Feature fuson of sde face and gat for vdeo-based human dentfcaton, Pattern Recognton 4, 008.

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1 4/14/011 Outlne Dscrmnatve classfers for mage recognton Wednesday, Aprl 13 Krsten Grauman UT-Austn Last tme: wndow-based generc obect detecton basc ppelne face detecton wth boostng as case study Today:

More information

Classifier Selection Based on Data Complexity Measures *

Classifier Selection Based on Data Complexity Measures * Classfer Selecton Based on Data Complexty Measures * Edth Hernández-Reyes, J.A. Carrasco-Ochoa, and J.Fco. Martínez-Trndad Natonal Insttute for Astrophyscs, Optcs and Electroncs, Lus Enrque Erro No.1 Sta.

More information

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

More information

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision SLAM Summer School 2006 Practcal 2: SLAM usng Monocular Vson Javer Cvera, Unversty of Zaragoza Andrew J. Davson, Imperal College London J.M.M Montel, Unversty of Zaragoza. josemar@unzar.es, jcvera@unzar.es,

More information

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Learning the Kernel Parameters in Kernel Minimum Distance Classifier Learnng the Kernel Parameters n Kernel Mnmum Dstance Classfer Daoqang Zhang 1,, Songcan Chen and Zh-Hua Zhou 1* 1 Natonal Laboratory for Novel Software Technology Nanjng Unversty, Nanjng 193, Chna Department

More information

Local Quaternary Patterns and Feature Local Quaternary Patterns

Local Quaternary Patterns and Feature Local Quaternary Patterns Local Quaternary Patterns and Feature Local Quaternary Patterns Jayu Gu and Chengjun Lu The Department of Computer Scence, New Jersey Insttute of Technology, Newark, NJ 0102, USA Abstract - Ths paper presents

More information

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification Introducton to Artfcal Intellgence V22.0472-001 Fall 2009 Lecture 24: Nearest-Neghbors & Support Vector Machnes Rob Fergus Dept of Computer Scence, Courant Insttute, NYU Sldes from Danel Yeung, John DeNero

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information

Edge Detection in Noisy Images Using the Support Vector Machines

Edge Detection in Noisy Images Using the Support Vector Machines Edge Detecton n Nosy Images Usng the Support Vector Machnes Hlaro Gómez-Moreno, Saturnno Maldonado-Bascón, Francsco López-Ferreras Sgnal Theory and Communcatons Department. Unversty of Alcalá Crta. Madrd-Barcelona

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

Discriminative classifiers for object classification. Last time

Discriminative classifiers for object classification. Last time Dscrmnatve classfers for object classfcaton Thursday, Nov 12 Krsten Grauman UT Austn Last tme Supervsed classfcaton Loss and rsk, kbayes rule Skn color detecton example Sldng ndo detecton Classfers, boostng

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

Recognition of Facial Expressions Based on Salient Geometric Features and Support Vector Machines

Recognition of Facial Expressions Based on Salient Geometric Features and Support Vector Machines The fnal publcaton s avalable at Sprnger va http://dx.do.org/10.1007/s11042-016-3428-9 Recognton of Facal Expressons Based on Salent Geometrc Features and Support Vector Machnes Deepa Ghmre 1, Joonwhoan

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers IOSR Journal of Electroncs and Communcaton Engneerng (IOSR-JECE) e-issn: 78-834,p- ISSN: 78-8735.Volume 9, Issue, Ver. IV (Mar - Apr. 04), PP 0-07 Content Based Image Retreval Usng -D Dscrete Wavelet wth

More information

Efficient Segmentation and Classification of Remote Sensing Image Using Local Self Similarity

Efficient Segmentation and Classification of Remote Sensing Image Using Local Self Similarity ISSN(Onlne): 2320-9801 ISSN (Prnt): 2320-9798 Internatonal Journal of Innovatve Research n Computer and Communcaton Engneerng (An ISO 3297: 2007 Certfed Organzaton) Vol.2, Specal Issue 1, March 2014 Proceedngs

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information

Learning a Class-Specific Dictionary for Facial Expression Recognition

Learning a Class-Specific Dictionary for Facial Expression Recognition BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 16, No 4 Sofa 016 Prnt ISSN: 1311-970; Onlne ISSN: 1314-4081 DOI: 10.1515/cat-016-0067 Learnng a Class-Specfc Dctonary for

More information

Human Face Recognition Using Generalized. Kernel Fisher Discriminant

Human Face Recognition Using Generalized. Kernel Fisher Discriminant Human Face Recognton Usng Generalzed Kernel Fsher Dscrmnant ng-yu Sun,2 De-Shuang Huang Ln Guo. Insttute of Intellgent Machnes, Chnese Academy of Scences, P.O.ox 30, Hefe, Anhu, Chna. 2. Department of

More information

An Image Fusion Approach Based on Segmentation Region

An Image Fusion Approach Based on Segmentation Region Rong Wang, L-Qun Gao, Shu Yang, Yu-Hua Cha, and Yan-Chun Lu An Image Fuson Approach Based On Segmentaton Regon An Image Fuson Approach Based on Segmentaton Regon Rong Wang, L-Qun Gao, Shu Yang 3, Yu-Hua

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Decson surface s a hyperplane (lne n 2D) n feature space (smlar to the Perceptron) Arguably, the most mportant recent dscovery n machne learnng In a nutshell: map the data to a predetermned

More information

Large-scale Web Video Event Classification by use of Fisher Vectors

Large-scale Web Video Event Classification by use of Fisher Vectors Large-scale Web Vdeo Event Classfcaton by use of Fsher Vectors Chen Sun and Ram Nevata Unversty of Southern Calforna, Insttute for Robotcs and Intellgent Systems Los Angeles, CA 90089, USA {chensun nevata}@usc.org

More information

Face Recognition using 3D Directional Corner Points

Face Recognition using 3D Directional Corner Points 2014 22nd Internatonal Conference on Pattern Recognton Face Recognton usng 3D Drectonal Corner Ponts Xun Yu, Yongsheng Gao School of Engneerng Grffth Unversty Nathan, QLD, Australa xun.yu@grffthun.edu.au,

More information

Smoothing Spline ANOVA for variable screening

Smoothing Spline ANOVA for variable screening Smoothng Splne ANOVA for varable screenng a useful tool for metamodels tranng and mult-objectve optmzaton L. Rcco, E. Rgon, A. Turco Outlne RSM Introducton Possble couplng Test case MOO MOO wth Game Theory

More information

CS 534: Computer Vision Model Fitting

CS 534: Computer Vision Model Fitting CS 534: Computer Vson Model Fttng Sprng 004 Ahmed Elgammal Dept of Computer Scence CS 534 Model Fttng - 1 Outlnes Model fttng s mportant Least-squares fttng Maxmum lkelhood estmaton MAP estmaton Robust

More information

Detection of an Object by using Principal Component Analysis

Detection of an Object by using Principal Component Analysis Detecton of an Object by usng Prncpal Component Analyss 1. G. Nagaven, 2. Dr. T. Sreenvasulu Reddy 1. M.Tech, Department of EEE, SVUCE, Trupath, Inda. 2. Assoc. Professor, Department of ECE, SVUCE, Trupath,

More information

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity Journal of Sgnal and Informaton Processng, 013, 4, 114-119 do:10.436/jsp.013.43b00 Publshed Onlne August 013 (http://www.scrp.org/journal/jsp) Corner-Based Image Algnment usng Pyramd Structure wth Gradent

More information

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task Proceedngs of NTCIR-6 Workshop Meetng, May 15-18, 2007, Tokyo, Japan Term Weghtng Classfcaton System Usng the Ch-square Statstc for the Classfcaton Subtask at NTCIR-6 Patent Retreval Task Kotaro Hashmoto

More information

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes SPH3UW Unt 7.3 Sphercal Concave Mrrors Page 1 of 1 Notes Physcs Tool box Concave Mrror If the reflectng surface takes place on the nner surface of the sphercal shape so that the centre of the mrror bulges

More information

Machine Learning 9. week

Machine Learning 9. week Machne Learnng 9. week Mappng Concept Radal Bass Functons (RBF) RBF Networks 1 Mappng It s probably the best scenaro for the classfcaton of two dataset s to separate them lnearly. As you see n the below

More information

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET 1 BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET TZU-CHENG CHUANG School of Electrcal and Computer Engneerng, Purdue Unversty, West Lafayette, Indana 47907 SAUL B. GELFAND School

More information

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros.

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros. Fttng & Matchng Lecture 4 Prof. Bregler Sldes from: S. Lazebnk, S. Setz, M. Pollefeys, A. Effros. How do we buld panorama? We need to match (algn) mages Matchng wth Features Detect feature ponts n both

More information

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

Determining the Optimal Bandwidth Based on Multi-criterion Fusion Proceedngs of 01 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 5 (01) (01) IACSIT Press, Sngapore Determnng the Optmal Bandwdth Based on Mult-crteron Fuson Ha-L Lang 1+, Xan-Mn

More information

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Face Recognition University at Buffalo CSE666 Lecture Slides Resources: Face Recognton Unversty at Buffalo CSE666 Lecture Sldes Resources: http://www.face-rec.org/algorthms/ Overvew of face recognton algorthms Correlaton - Pxel based correspondence between two face mages Structural

More information

Discriminative Dictionary Learning with Pairwise Constraints

Discriminative Dictionary Learning with Pairwise Constraints Dscrmnatve Dctonary Learnng wth Parwse Constrants Humn Guo Zhuoln Jang LARRY S. DAVIS UNIVERSITY OF MARYLAND Nov. 6 th, Outlne Introducton/motvaton Dctonary Learnng Dscrmnatve Dctonary Learnng wth Parwse

More information

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS P.G. Demdov Yaroslavl State Unversty Anatoly Ntn, Vladmr Khryashchev, Olga Stepanova, Igor Kostern EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS Yaroslavl, 2015 Eye

More information

Classifier Swarms for Human Detection in Infrared Imagery

Classifier Swarms for Human Detection in Infrared Imagery Classfer Swarms for Human Detecton n Infrared Imagery Yur Owechko, Swarup Medasan, and Narayan Srnvasa HRL Laboratores, LLC 3011 Malbu Canyon Road, Malbu, CA 90265 {owechko, smedasan, nsrnvasa}@hrl.com

More information

Lecture 4: Principal components

Lecture 4: Principal components /3/6 Lecture 4: Prncpal components 3..6 Multvarate lnear regresson MLR s optmal for the estmaton data...but poor for handlng collnear data Covarance matrx s not nvertble (large condton number) Robustness

More information

Pictures at an Exhibition

Pictures at an Exhibition 1 Pctures at an Exhbton Stephane Kwan and Karen Zhu Department of Electrcal Engneerng Stanford Unversty, Stanford, CA 9405 Emal: {skwan1, kyzhu}@stanford.edu Abstract An mage processng algorthm s desgned

More information

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration Improvement of Spatal Resoluton Usng BlockMatchng Based Moton Estmaton and Frame Integraton Danya Suga and Takayuk Hamamoto Graduate School of Engneerng, Tokyo Unversty of Scence, 6-3-1, Nuku, Katsuska-ku,

More information

Shape Representation Robust to the Sketching Order Using Distance Map and Direction Histogram

Shape Representation Robust to the Sketching Order Using Distance Map and Direction Histogram Shape Representaton Robust to the Sketchng Order Usng Dstance Map and Drecton Hstogram Department of Computer Scence Yonse Unversty Kwon Yun CONTENTS Revew Topc Proposed Method System Overvew Sketch Normalzaton

More information

Histogram of Template for Pedestrian Detection

Histogram of Template for Pedestrian Detection PAPER IEICE TRANS. FUNDAMENTALS/COMMUN./ELECTRON./INF. & SYST., VOL. E85-A/B/C/D, No. xx JANUARY 20xx Hstogram of Template for Pedestran Detecton Shaopeng Tang, Non Member, Satosh Goto Fellow Summary In

More information

Collaboratively Regularized Nearest Points for Set Based Recognition

Collaboratively Regularized Nearest Points for Set Based Recognition Academc Center for Computng and Meda Studes, Kyoto Unversty Collaboratvely Regularzed Nearest Ponts for Set Based Recognton Yang Wu, Mchhko Mnoh, Masayuk Mukunok Kyoto Unversty 9/1/013 BMVC 013 @ Brstol,

More information

Optimizing Document Scoring for Query Retrieval

Optimizing Document Scoring for Query Retrieval Optmzng Document Scorng for Query Retreval Brent Ellwen baellwe@cs.stanford.edu Abstract The goal of ths project was to automate the process of tunng a document query engne. Specfcally, I used machne learnng

More information

Multi-View Face Alignment Using 3D Shape Model for View Estimation

Multi-View Face Alignment Using 3D Shape Model for View Estimation Mult-Vew Face Algnment Usng 3D Shape Model for Vew Estmaton Yanchao Su 1, Hazhou A 1, Shhong Lao 1 Computer Scence and Technology Department, Tsnghua Unversty Core Technology Center, Omron Corporaton ahz@mal.tsnghua.edu.cn

More information

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

Heterogeneous Visual Features Fusion via Sparse Multimodal Machine

Heterogeneous Visual Features Fusion via Sparse Multimodal Machine 013 IEEE Conference on Computer Vson and Pattern Recognton Heterogeneous Vsual Features Fuson va Sparse Multmodal Machne Hua ang, Fepng Ne, Heng Huang, Chrs Dng Department of Electrcal Engneerng and Computer

More information

Face Recognition Based on SVM and 2DPCA

Face Recognition Based on SVM and 2DPCA Vol. 4, o. 3, September, 2011 Face Recognton Based on SVM and 2DPCA Tha Hoang Le, Len Bu Faculty of Informaton Technology, HCMC Unversty of Scence Faculty of Informaton Scences and Engneerng, Unversty

More information

Support Vector Machines. CS534 - Machine Learning

Support Vector Machines. CS534 - Machine Learning Support Vector Machnes CS534 - Machne Learnng Perceptron Revsted: Lnear Separators Bnar classfcaton can be veed as the task of separatng classes n feature space: b > 0 b 0 b < 0 f() sgn( b) Lnear Separators

More information

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 A mathematcal programmng approach to the analyss, desgn and

More information

2 ZHENG et al.: ASSOCIATING GROUPS OF PEOPLE (a) Ambgutes from person re dentfcaton n solaton (b) Assocatng groups of people may reduce ambgutes n mat

2 ZHENG et al.: ASSOCIATING GROUPS OF PEOPLE (a) Ambgutes from person re dentfcaton n solaton (b) Assocatng groups of people may reduce ambgutes n mat ZHENG et al.: ASSOCIATING GROUPS OF PEOPLE 1 Assocatng Groups of People We-Sh Zheng jason@dcs.qmul.ac.uk Shaogang Gong sgg@dcs.qmul.ac.uk Tao Xang txang@dcs.qmul.ac.uk School of EECS, Queen Mary Unversty

More information

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming CS 4/560 Desgn and Analyss of Algorthms Kent State Unversty Dept. of Math & Computer Scence LECT-6 Dynamc Programmng 2 Dynamc Programmng Dynamc Programmng, lke the dvde-and-conquer method, solves problems

More information

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour 6.854 Advanced Algorthms Petar Maymounkov Problem Set 11 (November 23, 2005) Wth: Benjamn Rossman, Oren Wemann, and Pouya Kheradpour Problem 1. We reduce vertex cover to MAX-SAT wth weghts, such that the

More information

A Modified Median Filter for the Removal of Impulse Noise Based on the Support Vector Machines

A Modified Median Filter for the Removal of Impulse Noise Based on the Support Vector Machines A Modfed Medan Flter for the Removal of Impulse Nose Based on the Support Vector Machnes H. GOMEZ-MORENO, S. MALDONADO-BASCON, F. LOPEZ-FERRERAS, M. UTRILLA- MANSO AND P. GIL-JIMENEZ Departamento de Teoría

More information

Image Matching Algorithm based on Feature-point and DAISY Descriptor

Image Matching Algorithm based on Feature-point and DAISY Descriptor JOURNAL OF MULTIMEDIA, VOL. 9, NO. 6, JUNE 2014 829 Image Matchng Algorthm based on Feature-pont and DAISY Descrptor L L School of Busness, Schuan Agrcultural Unversty, Schuan Dujanyan 611830, Chna Abstract

More information

Action Recognition Using Completed Local Binary Patterns and Multiple-class Boosting Classifier

Action Recognition Using Completed Local Binary Patterns and Multiple-class Boosting Classifier Acton Recognton Usng ompleted Local Bnary Patterns and Multple-class Boostng lassfer Yun Yang, Baochang Zhang, Lnln Yang School of Automaton Scence and Electrcal Engneerng Behang Unversty Beng, hna {yangyun,bczhang,yangln}@buaa.edu.cn

More information

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 48 CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 3.1 INTRODUCTION The raw mcroarray data s bascally an mage wth dfferent colors ndcatng hybrdzaton (Xue

More information

CS246: Mining Massive Datasets Jure Leskovec, Stanford University

CS246: Mining Massive Datasets Jure Leskovec, Stanford University CS46: Mnng Massve Datasets Jure Leskovec, Stanford Unversty http://cs46.stanford.edu /19/013 Jure Leskovec, Stanford CS46: Mnng Massve Datasets, http://cs46.stanford.edu Perceptron: y = sgn( x Ho to fnd

More information

A Gradient Difference based Technique for Video Text Detection

A Gradient Difference based Technique for Video Text Detection A Gradent Dfference based Technque for Vdeo Text Detecton Palaahnakote Shvakumara, Trung Quy Phan and Chew Lm Tan School of Computng, Natonal Unversty of Sngapore {shva, phanquyt, tancl }@comp.nus.edu.sg

More information

TN348: Openlab Module - Colocalization

TN348: Openlab Module - Colocalization TN348: Openlab Module - Colocalzaton Topc The Colocalzaton module provdes the faclty to vsualze and quantfy colocalzaton between pars of mages. The Colocalzaton wndow contans a prevew of the two mages

More information

Meta-heuristics for Multidimensional Knapsack Problems

Meta-heuristics for Multidimensional Knapsack Problems 2012 4th Internatonal Conference on Computer Research and Development IPCSIT vol.39 (2012) (2012) IACSIT Press, Sngapore Meta-heurstcs for Multdmensonal Knapsack Problems Zhbao Man + Computer Scence Department,

More information

Outline. Self-Organizing Maps (SOM) US Hebbian Learning, Cntd. The learning rule is Hebbian like:

Outline. Self-Organizing Maps (SOM) US Hebbian Learning, Cntd. The learning rule is Hebbian like: Self-Organzng Maps (SOM) Turgay İBRİKÇİ, PhD. Outlne Introducton Structures of SOM SOM Archtecture Neghborhoods SOM Algorthm Examples Summary 1 2 Unsupervsed Hebban Learnng US Hebban Learnng, Cntd 3 A

More information

Facial Expressions Recognition in a Single Static as well as Dynamic Facial Images Using Tracking and Probabilistic Neural Networks

Facial Expressions Recognition in a Single Static as well as Dynamic Facial Images Using Tracking and Probabilistic Neural Networks Facal Expressons Recognton n a Sngle Statc as well as Dynamc Facal Images Usng Trackng and Probablstc Neural Networks Had Seyedarab 1, Won-Sook Lee 2, Al Aghagolzadeh 1, and Sohrab Khanmohammad 1 1 Faculty

More information

Real-time Joint Tracking of a Hand Manipulating an Object from RGB-D Input

Real-time Joint Tracking of a Hand Manipulating an Object from RGB-D Input Real-tme Jont Tracng of a Hand Manpulatng an Object from RGB-D Input Srnath Srdhar 1 Franzsa Mueller 1 Mchael Zollhöfer 1 Dan Casas 1 Antt Oulasvrta 2 Chrstan Theobalt 1 1 Max Planc Insttute for Informatcs

More information

Video Object Tracking Based On Extended Active Shape Models With Color Information

Video Object Tracking Based On Extended Active Shape Models With Color Information CGIV'2002: he Frst Frst European Conference Colour on Colour n Graphcs, Imagng, and Vson Vdeo Object rackng Based On Extended Actve Shape Models Wth Color Informaton A. Koschan, S.K. Kang, J.K. Pak, B.

More information

Support Vector Machine for Remote Sensing image classification

Support Vector Machine for Remote Sensing image classification Support Vector Machne for Remote Sensng mage classfcaton Hela Elmanna #*, Mohamed Ans Loghmar #, Mohamed Saber Naceur #3 # Laboratore de Teledetecton et Systeme d nformatons a Reference spatale, Unversty

More information

A Gradient Difference based Technique for Video Text Detection

A Gradient Difference based Technique for Video Text Detection 2009 10th Internatonal Conference on Document Analyss and Recognton A Gradent Dfference based Technque for Vdeo Text Detecton Palaahnakote Shvakumara, Trung Quy Phan and Chew Lm Tan School of Computng,

More information

Facial Expression Recognition Based on Local Binary Patterns and Local Fisher Discriminant Analysis

Facial Expression Recognition Based on Local Binary Patterns and Local Fisher Discriminant Analysis WSEAS RANSACIONS on SIGNAL PROCESSING Shqng Zhang, Xaomng Zhao, Bcheng Le Facal Expresson Recognton Based on Local Bnary Patterns and Local Fsher Dscrmnant Analyss SHIQING ZHANG, XIAOMING ZHAO, BICHENG

More information

Object-Based Techniques for Image Retrieval

Object-Based Techniques for Image Retrieval 54 Zhang, Gao, & Luo Chapter VII Object-Based Technques for Image Retreval Y. J. Zhang, Tsnghua Unversty, Chna Y. Y. Gao, Tsnghua Unversty, Chna Y. Luo, Tsnghua Unversty, Chna ABSTRACT To overcome the

More information

Correlative features for the classification of textural images

Correlative features for the classification of textural images Correlatve features for the classfcaton of textural mages M A Turkova 1 and A V Gadel 1, 1 Samara Natonal Research Unversty, Moskovskoe Shosse 34, Samara, Russa, 443086 Image Processng Systems Insttute

More information

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning Computer Anmaton and Vsualsaton Lecture 4. Rggng / Sknnng Taku Komura Overvew Sknnng / Rggng Background knowledge Lnear Blendng How to decde weghts? Example-based Method Anatomcal models Sknnng Assume

More information

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures A Novel Adaptve Descrptor Algorthm for Ternary Pattern Textures Fahuan Hu 1,2, Guopng Lu 1 *, Zengwen Dong 1 1.School of Mechancal & Electrcal Engneerng, Nanchang Unversty, Nanchang, 330031, Chna; 2. School

More information

Classifying Acoustic Transient Signals Using Artificial Intelligence

Classifying Acoustic Transient Signals Using Artificial Intelligence Classfyng Acoustc Transent Sgnals Usng Artfcal Intellgence Steve Sutton, Unversty of North Carolna At Wlmngton (suttons@charter.net) Greg Huff, Unversty of North Carolna At Wlmngton (jgh7476@uncwl.edu)

More information

A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION

A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION 1 THE PUBLISHING HOUSE PROCEEDINGS OF THE ROMANIAN ACADEMY, Seres A, OF THE ROMANIAN ACADEMY Volume 4, Number 2/2003, pp.000-000 A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION Tudor BARBU Insttute

More information

Face Recognition Method Based on Within-class Clustering SVM

Face Recognition Method Based on Within-class Clustering SVM Face Recognton Method Based on Wthn-class Clusterng SVM Yan Wu, Xao Yao and Yng Xa Department of Computer Scence and Engneerng Tong Unversty Shangha, Chna Abstract - A face recognton method based on Wthn-class

More information

MOTION BLUR ESTIMATION AT CORNERS

MOTION BLUR ESTIMATION AT CORNERS Gacomo Boracch and Vncenzo Caglot Dpartmento d Elettronca e Informazone, Poltecnco d Mlano, Va Ponzo, 34/5-20133 MILANO boracch@elet.polm.t, caglot@elet.polm.t Keywords: Abstract: Pont Spread Functon Parameter

More information

Ecient Computation of the Most Probable Motion from Fuzzy. Moshe Ben-Ezra Shmuel Peleg Michael Werman. The Hebrew University of Jerusalem

Ecient Computation of the Most Probable Motion from Fuzzy. Moshe Ben-Ezra Shmuel Peleg Michael Werman. The Hebrew University of Jerusalem Ecent Computaton of the Most Probable Moton from Fuzzy Correspondences Moshe Ben-Ezra Shmuel Peleg Mchael Werman Insttute of Computer Scence The Hebrew Unversty of Jerusalem 91904 Jerusalem, Israel Emal:

More information

Gender Classification using Interlaced Derivative Patterns

Gender Classification using Interlaced Derivative Patterns Gender Classfcaton usng Interlaced Dervatve Patterns Author Shobernejad, Ameneh, Gao, Yongsheng Publshed 2 Conference Ttle Proceedngs of the 2th Internatonal Conference on Pattern Recognton (ICPR 2) DOI

More information

AP PHYSICS B 2008 SCORING GUIDELINES

AP PHYSICS B 2008 SCORING GUIDELINES AP PHYSICS B 2008 SCORING GUIDELINES General Notes About 2008 AP Physcs Scorng Gudelnes 1. The solutons contan the most common method of solvng the free-response questons and the allocaton of ponts for

More information

Title: A Novel Protocol for Accuracy Assessment in Classification of Very High Resolution Images

Title: A Novel Protocol for Accuracy Assessment in Classification of Very High Resolution Images 2009 IEEE. Personal use of ths materal s permtted. Permsson from IEEE must be obtaned for all other uses, n any current or future meda, ncludng reprntng/republshng ths materal for advertsng or promotonal

More information

Incremental Multiple Kernel Learning for Object Recognition

Incremental Multiple Kernel Learning for Object Recognition Incremental Multple Kernel Learnng for Obect Recognton Anruddha Kembhav, Behat Sddque, Roland Mezano, Scott McClosey, Larry S. Davs Unversty of Maryland, College Par Honeywell Labs Abstract A good tranng

More information

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur FEATURE EXTRACTION Dr. K.Vjayarekha Assocate Dean School of Electrcal and Electroncs Engneerng SASTRA Unversty, Thanjavur613 41 Jont Intatve of IITs and IISc Funded by MHRD Page 1 of 8 Table of Contents

More information

Fast Feature Value Searching for Face Detection

Fast Feature Value Searching for Face Detection Vol., No. 2 Computer and Informaton Scence Fast Feature Value Searchng for Face Detecton Yunyang Yan Department of Computer Engneerng Huayn Insttute of Technology Hua an 22300, Chna E-mal: areyyyke@63.com

More information

Modular PCA Face Recognition Based on Weighted Average

Modular PCA Face Recognition Based on Weighted Average odern Appled Scence odular PCA Face Recognton Based on Weghted Average Chengmao Han (Correspondng author) Department of athematcs, Lny Normal Unversty Lny 76005, Chna E-mal: hanchengmao@163.com Abstract

More information

Image Alignment CSC 767

Image Alignment CSC 767 Image Algnment CSC 767 Image algnment Image from http://graphcs.cs.cmu.edu/courses/15-463/2010_fall/ Image algnment: Applcatons Panorama sttchng Image algnment: Applcatons Recognton of object nstances

More information

SUMMARY... I TABLE OF CONTENTS...II INTRODUCTION...

SUMMARY... I TABLE OF CONTENTS...II INTRODUCTION... Summary A follow-the-leader robot system s mplemented usng Dscrete-Event Supervsory Control methods. The system conssts of three robots, a leader and two followers. The dea s to get the two followers to

More information

Biostatistics 615/815

Biostatistics 615/815 The E-M Algorthm Bostatstcs 615/815 Lecture 17 Last Lecture: The Smplex Method General method for optmzaton Makes few assumptons about functon Crawls towards mnmum Some recommendatons Multple startng ponts

More information

Network Intrusion Detection Based on PSO-SVM

Network Intrusion Detection Based on PSO-SVM TELKOMNIKA Indonesan Journal of Electrcal Engneerng Vol.1, No., February 014, pp. 150 ~ 1508 DOI: http://dx.do.org/10.11591/telkomnka.v1.386 150 Network Intruson Detecton Based on PSO-SVM Changsheng Xang*

More information

Face Tracking Using Motion-Guided Dynamic Template Matching

Face Tracking Using Motion-Guided Dynamic Template Matching ACCV2002: The 5th Asan Conference on Computer Vson, 23--25 January 2002, Melbourne, Australa. Face Trackng Usng Moton-Guded Dynamc Template Matchng Lang Wang, Tenu Tan, Wemng Hu atonal Laboratory of Pattern

More information

Face Detection with Deep Learning

Face Detection with Deep Learning Face Detecton wth Deep Learnng Yu Shen Yus122@ucsd.edu A13227146 Kuan-We Chen kuc010@ucsd.edu A99045121 Yzhou Hao y3hao@ucsd.edu A98017773 Mn Hsuan Wu mhwu@ucsd.edu A92424998 Abstract The project here

More information

Robust Face Alignment for Illumination and Pose Invariant Face Recognition

Robust Face Alignment for Illumination and Pose Invariant Face Recognition Robust Face Algnment for Illumnaton and Pose Invarant Face Recognton Fath Kahraman 1, Bnnur Kurt 2, Muhttn Gökmen 2 Istanbul Techncal Unversty, 1 Informatcs Insttute, 2 Computer Engneerng Department 34469

More information

Categorizing objects: of appearance

Categorizing objects: of appearance Categorzng objects: global and part-based models of appearance UT Austn Generc categorzaton problem 1 Challenges: robustness Realstc scenes are crowded, cluttered, have overlappng objects. Generc category

More information

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance Tsnghua Unversty at TAC 2009: Summarzng Mult-documents by Informaton Dstance Chong Long, Mnle Huang, Xaoyan Zhu State Key Laboratory of Intellgent Technology and Systems, Tsnghua Natonal Laboratory for

More information

Solving two-person zero-sum game by Matlab

Solving two-person zero-sum game by Matlab Appled Mechancs and Materals Onlne: 2011-02-02 ISSN: 1662-7482, Vols. 50-51, pp 262-265 do:10.4028/www.scentfc.net/amm.50-51.262 2011 Trans Tech Publcatons, Swtzerland Solvng two-person zero-sum game by

More information

MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS XUNYU PAN

MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS XUNYU PAN MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS by XUNYU PAN (Under the Drecton of Suchendra M. Bhandarkar) ABSTRACT In modern tmes, more and more

More information

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010 Smulaton: Solvng Dynamc Models ABE 5646 Week Chapter 2, Sprng 200 Week Descrpton Readng Materal Mar 5- Mar 9 Evaluatng [Crop] Models Comparng a model wth data - Graphcal, errors - Measures of agreement

More information

Announcements. Supervised Learning

Announcements. Supervised Learning Announcements See Chapter 5 of Duda, Hart, and Stork. Tutoral by Burge lnked to on web page. Supervsed Learnng Classfcaton wth labeled eamples. Images vectors n hgh-d space. Supervsed Learnng Labeled eamples

More information

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 1, JANUARY

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 1, JANUARY IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 1, JANUARY 2015 189 Dscrmnatve Shared Gaussan Processes for Multvew and Vew-Invarant Facal Expresson Recognton Stefanos Eleftherads, Student Member,

More information