Cast Indexing for Videos by NCuts and Page Ranking

Size: px
Start display at page:

Download "Cast Indexing for Videos by NCuts and Page Ranking"

Transcription

1 Cast Indexng or Vdeos by NCuts and Page Rankng Yong Gao, Tao Wang, Janguo L, YangZhou Du, We Hu, Ymn Zhang, * HaZhou A Intel Chna Research Center, Beng, P.R. Chna * Det. CS. &Tech. o Tsnghua Unversty, Beng, P.R. Chna { yong.y.gao, tao.wang, anguo.l, yangzhou.du we.hu, ymn.zhang }@ntel.com ahz@mal.tsnghua.edu.cn ABSTRACT Cast ndexng s an mortant vdeo mnng technque whch rovdes audence the caablty to ecently retreve nterested scenes, events, and stores rom a long vdeo. Ths aer rooses a novel cast ndexng aroach based on Normalzed Grah Cuts NCuts) and Page Rankng. The system rst adots ace tracker to grou ace mages n each shot nto ace sets, and then extract local SIFT eature as the eature reresentaton. There are two key roblems or cast ndexng. One s to nd an otmal artton to cluster ace sets nto man cast. The other s how to exlot the latent relatonshs among characters to rovde a more accurate cast rankng. For the rst roblem, we model each ace set as a grah node, and adot Normalzed Grah Cuts NCuts) to realze an otmal grah artton. A novel local neghborhood dstance s roosed to measure the dstance between ace sets or NCuts, whch s robust to outlers. For the second roblem, we buld a relaton grah or characters by ther co-occurrence normaton, and then adot the PageRank algorthm to estmate the Imortant Factor IF) o each character. The PageRank IF s used wth the content based retreval score or nal rankng. Extensve exerments are carred out on moves, TV seres and home vdeos. Promsng results demonstrate the eectveness o roosed methods. Categores and Subect Descrtors H.3.1 [Inormaton Storage and Retreval]: Content Analyss and Indexng. General Terms Algorthms, Management, Desgn, Exermentaton. Keywords Cast Indexng, Man Cast Detecton, Cast Rankng, Local Neghbor Dstance, NCuts, Page Rankng. 1. INTRODUCTION The ongong exanson o vdeo data n World Wde Web, move, Permsson to make dgtal or hard coes o all or art o ths work or ersonal or classroom use s granted wthout ee rovded that coes are not made or dstrbuted or rot or commercal advantage and that coes bear ths notce and the ull ctaton on the rst age. To coy otherwse, to reublsh, to ost on servers or to redstrbute to lsts, requres ror secc ermsson and/or a ee. CIVR'07, July 9-11, 2007, Amsterdam, The Netherlands. Coyrght 2007 ACM /07/ $5.00. TV rograms, and home vdeos has generated ncreasng requrements or semantc based vdeo mnng technques, such as news abstracton, sorts hghlghts detecton, move/tv/home vdeos ndexng and retreval etc. As the man subect o scenes, events and stores n these vdeos, characters are one o the most mortant contents to be ndexed. By cast ndexng, we can dscover actve characters who requently aear n the vdeo and retreve ther assocated vdeo cls or ecent browsng. For cast ndexng, human ace s one o the most mortant vsual cues besdes auxlary cues o seech, and clothes etc. Automatc ace detecton and recognton technques can be emloyed as man ways and means or cast ndexng. However aces n vdeos, esecally lms, stcoms, and home vdeos, usually have large varatons o ose, exresson and llumnaton whch ust are the reasons why relable ace recognton s stll a very challengng roblem n ractce[11]. To reduce the adverse eect o those varatons n mage/vdeo based ace recognton, a lot o methods have been roosed. Arandelovc and Zssserman roose to aly ane warng and llumnaton correctng on ace mages to allevate the adverse eects nduced by ose and llumnaton varatons [2]. But t s unable to deal wth the out-o-lan ace rotaton. In [1] and [7], ace recognton s based on manold analyss. However the real manolds o aces and relatonshs among them n vdeos are too comlex to be accurately characterzed by smled models. Although 3D ace model s utlzed to enhance the vdeo-based ace recognton erormance [3], t s very dcult to accurately recover the head ose arameters by the state-o-the-art regstraton technques and not ractcal or real-world alcaton. In a word, t s very hard to buld a robust cast ndexng system only based on ace recognton technques. Fortunately, n eature-length lms, e.g., move and TV seres, the man characters requently aear n derent shots, resultng n large numbers o contnuous aces. The rch dynamc acal normaton rovdes mult-vew ace exemlars o the same erson, whch make t ossble to detect man characters by clusterng aces. Accordng to the dea, ths aer rooses a novel case ndexng aroach. Frst, near rontal aces are detected and tracked nto ace sets by ace tracker and local SIFT eatures are extracted on normalzed aces. Then we use ace set as the basc rocessng unt to detect man characters that s essentally a attern clusterng roblem. To reduce adverse eects o ace outlers, a local neghbor dstance s roosed to measure the smlarty between ace sets. By modelng ace sets as nodes o a grah, Normalzed Grah CutsNCuts) s emloyed to nd the otmal artton. The NCuts based clusterng results rovde a

2 Shot and Scene detecton Face trackng Landmark onts detecton Local SIFT eature extracton Feature extracton module Dstance between ace sets Man cast detecton by NCuts Man cast rankng by IF Cast ndexng module Fgure 1. Framework o the roosed cast ndexng system content-based browsng result, but not rank how mortant each character s. Motvated by the successul age rankng technology o Google web search [6], we roose a novel character mortant actor IF) aroach to rank cast lst, whch s a used score o Page Rankng, aearng tme and requency. By age rankng, we can also dscover the latent relatonshs between characters. The aer s organzed as ollows. In Secton 2, the cast ndexng system and eature extracton module are rst ntroduced. Then we roose an mroved dstance measure and the NCuts based man cast detecton aroach. To rank characters by ther mortance actor, a cast rankng method s roosed. Fnally, extensve exermental results are reorted n Secton 3, ollowed by the conclusons n Secton SYSTEM FRAMEWORK The cast ndexng system conssts o three man modules as llustrated n Fgure 1. In the eature-extracton module, shot boundary and scene segmentaton are rst detected. Then near rontal aces are detected and tracked by ace tracker n each shot. At last, local SIFT eatures are extracted on normalzed ace mages. In the man cast detecton module, man characters are detected by usng NCuts algorthm to cluster ace set. Fnally, or ecent browsng, cast rankng module sorts characters by ther mortance actor IF). 2.1 FEATURE EXTRACTION MODULE Smlar to text rocessng based on word, sentence, aragrah and document, vdeo mnng can be analyzed rom our levels,.e., rame, shot, scene and the whole vdeo sequence level. A shot s a set o vdeo rames catured by a sngle camera n one consecutve recordng acton. A scene s dened as one o the subdvsons o a vdeo n whch the settng s xed and tme contnuous, or resentng the contnuous acton n one lace. Our system uses a shot detecton algorthm [4] to detect shots. Based on detected shots, we cluster temoral-satal coherent shots nto scenes by NCuts algorthm and color hstogram eatures[10]. In each shot, we use ace tracker [14] to detect and track near rontal aces and outut ace sets. As the basc rocessng unt o cast ndexng, ace sets not only rovde ror knowledge about mult-vew acal exemlars whch belong to the same erson but also decrease the data sze o ace clusterng algorthm n man cast detecton. For each ace, actve shae model ASM) based ace algnment algorthm [8] s utlzed to detect 88 acal landmarks as shown n g. 2b). Usng these landmarks, ace mages are geometrcally normalzed nto a standard orm by ane transormaton to remove the varatons o translaton, scale, n-lane and slght out-o-lane rotaton. It s demonstrated that local eatures outerorm global ones n most recognton and vercaton tasks because they are more robust to artal occlusons, ose and llumnaton varatons. We extract the SIFT eatures [9] on ve local regons whch coverng two eyes, central regons o two eyes, nose and orehead, as shown n Fgure 2c). 2.2 DISTANCE MEASURE The ace tracker grous same character s aces n each shot nto ace sets. However, characters may aears n multle shots, and hence n multle ace sets. As llustrated n Fgure 3, the am o man cast detecton s to cluster ace sets nto bgger aggregaton whch ncludes all the ace sets o the same character. For ace set clusterng, dstance measure between two ace sets s a key ssue. a) b) c) Fgure 2. Local SIFT eature extracton. a) The orgnal ace mage; b) Detected 88 landmarks by ace algnment; c) Fve local regons or SIFT eature extracton. Fgure 3. Man cast detecton by clusterng ace sets

3 The ollowng asects should be consdered: 1) The number o ace mages n ace sets s derent, whch are relevant to the ace aearng tme n derent shot. I the ose, exresson and llumnaton o aces dynamcally changes n a ace set, they can rovde rch mult-vew exemlars o the same erson, whch s very useul to brdge ace sets wth overlaed ace exemlars or ace clusterng. On the contrary, aeared aces n a ace set are statc or the ace number s very small, t wll be not very normatve. 2) The manolds o ace sets n eature sace are very comlex and qute derent rom each other. Fgure 4 shows manolds o our ace sets o two characters n the oular Korea TV seres DaChangJn. It can be observed that aces o the same erson can be very dstant, whle derent erson s aces may be very near n the eature sace. However, two ace sets wth bgger overla generally belong to the same erson wth hgher robablty. 3) Unavodably, there are outlers due to msalgnment and varatons o ose, exresson or llumnaton etc. When clusterng ace sets, t wll roagate the error and merge ace sets o derent ersons nto the same cluster. Fortunately, there s rch acal no n eather length lms. Snce outlers don t occurred requently, aces wth hgher densty dstrbuton n the eature sace relably belong to same erson. We have one undament n mnd: normal samles have suort rom ther nearest neghborhood same-ace-set Face-set 1 Face-set 2 Face-set 3 Face-set 4 a) samles, whle outlers do not Dstance Measure between Two Faces Based on the above observaton 3), we dene dstance measure between two ace mages by consderng ther nearest neghborhood suort normaton. In ths aer, k-nearest neghbor s adoted. Let S and S are two ace sets, or two aces xm S and x n S, the local neghbor dstance between x m and x n s dened as: 1 d xm, x n) = x x 2 q 1) k x Ν xm ), x q Ν x n ) where Ν x m ) s the k neghbors o x m n S and Ν x n ) s the k neghbors o x n n S,. denotes L 2 dstance between two aces. It can be easly roved: d x, x ) x x 2) m n q x Ν xm ) x q Ν x n ) It s to say that the dstance measure dened by Eq1) s equvalent to rst aly smooth lter on manold to weaken or remove outler dsturbances, then calculated dstance between two averaged data onts. As llustrated n g. 5, the local neghbor dstance uncton o Eq1) s more robust to outler than smle L2 dstance x x. m n Dstance Measure between Two Face Sets As mentoned n the observaton 2), the ace sets wth bgger overla generally belong to the same erson wth hgher robablty. For the dstance measure between two ace sets, t s ntutve to summarze l mnmum local neghbor dstances d x m, x ) to evaluate the ace set overla. n d S, S ) = l l x S, x S m mn n d x m, x Where mn l s the l th mnmum dstance o d x m, x ), n )) n 3) dr 1,g 1 ) ace set G g 1 ace set R r 1 dr 1,b 1 ) ace set B b 1 b) Fgure 4. Comlexty o the ace set manolds n eaturelength lms. a) Reresentatve ace mages o our ace sets o two characters. b) Corresondng manolds are vsualzed by rst two dmensons o PCA subsace. Fgure 5. Dstance measure between two aces that comes rom two ace sets resectvely. The cross onts are the averaged onts o local neghbor onts o r 1, g 1, and b 1 resectvely. Although green onts g 1 s more near to red ont than blue onts b 1, the local neghbor dstance dr 1, g 1 )>dr 1, b 1 ) snce g 1 onts has lower local densty than b 1 onts.

4 xm S and x n S. I there s an enough overla between ace set S and S, or the two ace sets are near enough searately, the summary o mn l d xm, x n )) wll be small. And ace set S and S belong to the same erson wth hgh robablty. Otherwse, we won t make decson to merge the two ace sets untl they are brdged by other neghbored ace sets. In summary, emloyng local neghbor dstance elmnates the outler dsturbance and acheves more robust erormance to measure the smlarty between ace sets as llustrated n g MAIN CAST DETECTION Ater the dstance measure between ace sets s dened, the man cast detecton s modeled as a grah arttonng roblem,.e. grah cut. All ace sets are reresented as a weghted undrected grah G = V, E), where the nodes V o the grah are the ace sets and the edges are the smlartes between ar-wse ace sets. For ace set clusterng, we seek the best artton C, C, 1 2 K, C to m satsy that the smlarty among the nodes n a sub-grah C s hgh and across smlarty between sub-grahs C, C ) s low. To otmally artton a grah consttuted by ace sets, the normalzed cut algorthm NCuts) s emloyed Normalzed Grah Cuts The normalzed cut s rst roosed by Sh and Malk [5]. Here we gve a bre ntroducton. A grah G = V, E) can be arttoned nto two dsont sub-grahs A and B wth A B = V and A B = Φ, by smly removng edges connectng the two arts. The degree o dssmlarty between these two sub-grahs can be comuted as total weght o the edges that have been removed. In grah theoretc language, t s called the cut: cut A, = w u, v) 4) u A, v B The otmal bartton o a grah s the one that mnmze ths cut value. To avod unnatural bas or arttonng o small sets o onts, Sh and Malk roosed the dsassocaton measure o the normalzed cut Ncut) [5]: cut A, cut A, Ncut A, + assoc A, V ) assoc B, V ) = 5) where assoc A, V ) = u A t w u, t) s the total connecton rom, V nodes n A to all nodes n the grah and assoc B, V ) s smlarly dened. Gven a artton o the grah,.e., dvde V nto two dsont sets A and B, let X be a N = V dmensonal ndcaton vector, x = 1 node s n A and -1 node s n B. Let d ) = w, ) and D be and N N dagonal matrx wth d on ts dagonal, W be an N N symmetrcal matrx wth W, ) = w, the aroxmate dscrete soluton to mnmzng NCuts can be ound ecently by solvng the generalzed egenvalue system, D W ) Y = λdy 6) where Y s a lnear transormaton o X and can be used or artton by a threshold The Man Cast Detecton Algorthm For two ace sets S and S, the grah edge weght w s dened as: 2 d S, S ) 2 σ w = e S s the n nearest neghbor o S 7) 0 otherwse In our exerments, n s set to the square root o the number o ace sets N and σ s set to 0.8 whch s about the threshold that two aces are rom the same character n SIFT eature sace. Based on NCuts clusterng aroach, the man cast detecton algorthm conssts o the ollowng stes: 1. Gven ace sets detected n the eature extracton module, set u a weghted grah G = V, E) usng dstance uncton dened by Equaton 2) and 4). 2. From the grah, create matrx W and D. And solve egenvalue system D W ) x = λdx. 3. Use the egenvector wth the second smallest egenvalue to bartton the grah by ndng the slttng ont wth mnmum Ncut. 4. Recursvely aartton the sub-grah when stong crteron s not satsed. Whether contnue to bartton a sub-grah s decded by tryng a new bartton. A sub-grah should be arttoned ether o the ollowng two condtons are satsed: 1. The NcutA, o the tryng bartton s below a reselected value. 2. Comutng the hstogram o the egenvector values and the rato between the mnmum and the maxmum values n the bns s not smaller than a re-selected threshold. 2.4 MAIN CAST RANKING In Secton 2.3, the man cast detecton module oututs ace set clusters o requently aeared characters. To sort the mortant characters and analyze ther relatonshs n scenes, we urther rank man casts clusters o ace sets) by Imortance Factor IF). It s natural that mortant actors generally aear wth hgh aearng tme or aearng requency. Furthermore, as the hnge o story scenaro, mortant actors requently aear wth others n scenes, e.g. dalog, ghtng etc. Thus the character mortance and ther close relatonshs can be dscovered rom ther ont aearng requency n scenes. Motvated by the successul PageRank technology o Google web search [6], each character can be vewed as a web age and ontly aeared multle characters n a scene can be vewed as lnked edges among them. It s ntutve that one character has many connected edges wth others, or t s connected to some mortant ersons, the PageRank value s bg. In our system, we rank the detected characters by Imortance Factor IF), whch uses scores o PageRank [12], aearng tme and aearng requency by a lnear weghted average. For a character C, hs/her IF can be calculated by the ollowng ormula

5 IF C ) wt At C ) + w A C ) + w A C ) = 8) where At ) s the Aearng Tme AT) score, A ) s the Aearng Frequency AF) score and A ) s the PageRank score. In our exerments, w t = 0. 2, w = 0. 3, w = 0. 5 are ther corresondng weghts. The PageRank score s calculated usng the algorthm roosed by Lawrence Page and Sergey Brn [12]. For characters C, =1,2..N, the PageRank value o C s dened as the ollows: A C ) = 1 d) + d A C ) / L C ), =1,..,N 9) where A C ) s the PageRank score o the character C, A C ) s the PageRank score o character C whch lnks to character C,.e. C ontly aear wth C n a same scene. L C ) s the outbound lnks o C,.e., the number o characters who ontly aeared wth C. d s a damng actor n [0, 1]. Each character s assgned a startng PageRank value A C ) =1 and the damng actor s set to 0.8. Then an teratve scheme s emloyed to calculate the PageRank scores o each character. Lawrence Page and Sergey Brn have roved that the PageRank value wll converge to the real rank values no matter what the ntal rank values A C ) are. The aearng tme At C ) and the aearng requency A C ) scores are calculated accordng to the character aearng tme and the clustered ace set number. In detal, the two scores are Shots & Scenes Face sets Scene1 Characters & Relatonshs Cast rankng Scene2 Cr Scene3 C r > C b > C g Scene4 Scene5 Fgure 6 An llustratve examle o three characters C r, C g, and C b aeared n a vdeo wth 6 scenes. The undrected edge reresents two ace sets ontly aeared n a scene. The ellse sze reresents the aearng tme. Table 1 Imortance Factors o each character n g. 5. It can be observed that Cr has the hghest IF score and s detected as the most mortant character. Character Cr Cg Cb AtC) AC) LC) AC) IFC) Cg Cb Scene6 dened as ollows. A C ) A t C ) = C ' s aearng tme C ' s aearng tme = 10) C ' s ace set number 11) C ' s ace set number The aearng tme and aearng requency scores are useul actors when we udge whether a character s mortant by hmsel. And the PageRank scores relect the mortance o character relatonshs. The character that aears wth more other characters and more mortant characters wll get hgher PageRank score. By ths way, the character relatonshs can be drectly got by the PageRank analyss. An llustratve examle s shown n Fgure 6. The calculated IF scores are lsted n table 1. It can be observed that C r s more mortant than Cg Snce Cr has more relatonshs wth others than Cg. Although Cb aears wth the longest tme AtCb)=0.6, t s not mortant due to bad aearng requency score ACb)=0.2 and age rank score ACb) = EXPERIMENTS To evaluate erormance o the roosed cast ndexng aroach, we conducted exerments on our vdeos whch are one Hollywood move 007 De Another Day 007), two TV seres DaChangJn DCJ), rends FR), and one home vdeoshv). For each vdeo, two ersons are manually annotated wth ground truth. The detals are lsted n Table EVALUATION METRIC ~ The normalzed mutual normaton I X, Y ) [12] s utlzed to evaluate the man cast detecton erormance. Based on normaton theory, mutual normaton s a deendence measure or two random varables whch s dened as: I x, y) X, Y ) = x, y)log 12) x X y Y x) y) The normalzed mutual normaton s calculated by: ~ I X, Y ) I X, Y ) = 13) max H X ), H Y )) where H X ) and H Y ) are the entroy o the cluster sets X and Y resectvely. It s obvous that I ~ s n the range o [0,1]. The more smlar the dstrbuton between X and Y s, the bgger ~ I X, Y ) s. Thus the bgger value I ~ X, Y ) means the better clusterng erormance. Table 2 Statstc o our testng vdeos Vdeo name 007 DCJ FR HV Vdeo length 4022s 2680s 1365s 2885s Shots number Scenes number Face Set Characters

6 3.2 MAIN CAST DETECTION RESULTS Fgure 7 llustrates the detected man cast lst and manually annotated ones o the our test vdeos, where the mssng man casts, non-man-casts and dulcate casts are labeled wth M, N and D resectvely. From the gure, t can be observed that the cast ndexng system can dscover the man cast o the our vdeos well totally 39 man characters, mssng 5) wth very low alse detecton rate only 2 non-man-casts are detected). One man roblem s the dulcate characters due to bg varatons o ace ose, exresson and llumnaton etc. We beleve that the dulcate case can be reduced by better ace algnment rerocessng technques. We also quanttatvely evaluate the erormance o the man cast detecton usng mutual normaton by comarng automatcally clusterng results wth manually labeled ace sets. For comarson, several other methods are also tested. Method 1: The roosed method n ths aer. Method 2: Usng NCuts to cluster ace sets and mnmum ace ar dstance between two ace sets, whch s dened as d S, S ) = mn x x 14) xm S, x n S where. denotes L 2 dstance between two aces n eature sace. Method 3: Usng HAC Herarchcal Agglomeratve Clusterng) to cluster ace sets and mnmum ace ar dstance between two ace sets Equaton 14) to measure the smlarty between them. Ths s our revous method [13]. m n Fgure 8. Mutual Inormaton o the our methods on the our test vdeos Fgure 7. Detected man cast o our vdeos. The cast lsts wth lght-green background are manually annotated ground truth. Blue box wth character M ndcates the mssng detected casts. Green box wth character D ndcates dulcate casts. Red box wth character N ndcates casts whch are not the man casts. Fgure 9. Ranked cast lst accordng derent methods. a) Manually ranked cast lst. b) Cast lst ranked by the roosed method. c) Cast lst ranked by aearng tme. d) Cast lst ranked by aearng requency.

7 Method 4: Usng HAC to cluster ace sets and local neghbor dstance roosed n ths aer to measure the smlarty between two ace sets. The mutual normaton values o these methods on derent vdeos are shown n Fgure 8. It s obvous that the roosed one NCuts+Local neghbor dstance) has the best erormance among the our clusterng methods. And the roosed local neghbor dstance can actually mrove the clusterng erormance or both NCuts and HAC. Excet on the vdeo 007 De Another Day, NCuts acheves much better erormance than HAC. One man roblem o HAC s that t roduces many solated data onts aggregaton o ace sets) or lower threshold whle t mxes data onts o derent classes or hgher threshold. Another observaton s that the erormance, measured by mutual normaton, vares greatly on derent vdeos. It acheves the best erormance on oular TV seres DaChangJn and Frends wth MI and the worst on home vdeo XAn wth MI Ths can be exlaned rom the asects o qualty o ace cture, shootng style and content o the story, whch wll greatly aect the cture qualty, ose and llumnaton varaton o ace mages n the vdeos. 3.3 MAIN CAST RANKING RESULTS Ater the man cast detecton, every man cast s IF s calculated accordng Equaton 8-11 and the cast lst s ranked accordng to the IF scores. Fgure 9 shows the ranked cast lsts o move 007 De Another Day sortng by derent schemes. In Fgure 9, a) s the manually ranked cast lst based the mortance o derent casts. Sub-Fgure b) s the ranked cast lst based on IF scores roosed n the aer. c) and d) are casts lsts ranked based on aearng tme and aearng requency. I we examne the lsts careully, t can be ound that the order o cast lst n Fgure 9 b) s more close to that o cast lst n Fgure 9 a). That s, the IF scores can relect the mortance o man casts more reasonably than aearng tme or aearng requency. 4. CONCLUSION Based on NCuts and age rankng, ths aer rooses a novel cast ndexng aroach or eature length lms. Gven an nut vdeo, ace tracker rst tracks aces rom each shot, and oututs ace sets. Then by clusterng smlar/overlaed ace sets, NCuts s emloyed to detect man characters. To deal wth outler aces rom varatons o ose, exresson, llumnaton and bad ace normalzaton, we use local neghbor dstance to measure the smlarty between ace sets. To nd the best artton on ace sets, normalzed cut s emloyed as a crteron to artton an undrected grah n whch every ace set s reresented by a node and the otmal soluton can be easly sought by solvng egenvalue roblem. Furthermore, ater man character detecton, a cast rankng emrcal ormula s roosed to rank characters and mnng ther relatonshs. Extensve exerments demonstrate that the NCuts and age rankng based cast ndexng aroach s a rutul drecton to exlore robust cast ndexng system. 5. REFERENCES [1] Arandelovc, O., Shakhnarovch, G., Fsher, J., Colla, R. and Darrell, T. Face Recognton wth Image Sets Usng Manold Densty Dvergence. In Proceedngs o the 2005 IEEE Comuter Socety Con. on CVPR. 2005, [2] Arandelovc, O. and Zsserman, A. Automatc Face Recognton or Flm Character Retreval n Feature-Length Flms. In Proceedngs o the 2005 IEEE Comuter Socety Con. on CVPR. 2005, [3] Everngham, M. and Zsserman, A. Identyng Indvduals n Vdeo by Combnng Generatve and Dscrmnatve Head Models. In Proceedngs o the Tenth IEEE Int. Con. On Comuter Vson. 2005, [4] Yuan J.H., Zheng W.J., Chen L., et. al. Tsnghua Unversty a TRECVID 2004: shot boundary detecton and hgh-level eature extracton, In NIST worksho o TRECVID [5] Janbo, S. and Malk, J. Normalzed Cuts and Image Segmentaton. IEEE Trans. On PAMI, vol 22, no. 8, 2000, [6] Lawrence, P. and Sergey, B. The PageRank Ctaton Rankng: Brngng Order to the Web. Stanord Dgtal Lbrary Technologes Proect, 1998 [7] Lee, K. C. and Kregman, D. Onlne Learnng o Probablstc Aearng Manolds or Vdeo-based Recognton and Trackng. In Proceedngs o the 2005 IEEE Comuter Socety Con. on CVPR. 2005, [8] Zhang L., A HZ, Lao SH, Robust Face Algnment Based On Herarchcal Classer Network, T.S. Huang et al. Eds.): HCI/ECCV 2006, LNCS 3979,.1-11, Srnger- Verlag Berln Hedelberg [9] Lowe, D. Dstnctve Image Features From Scale-Invarant Keyonts. Internatonal Journal o Comuter Vson. 2004, [10] Rasheed, Z., Shah, M., Detecton and reresentaton o scenes n vdeos. IEEE Transactons on Multmeda, Volume 7, Issue 6, Dec. 2005, [11] Stan Z., L., Jan, A. K. Handbook o Face Recognton. Srnger, NY, 2004 [12] Studholme, C., Hll, D. and Hawkes, D. An overla nvarant entroy measure o 3D medcal mage algnment. Pattern recognton. 1999, [13] Fan W., Wang T., Bouguet, J. Y., Hu W., et. al. Semsuervsed Cast Indexng or Feature-Length Flms. In Proeedngs o the 13 th Internatonal MultMeda Modellng Conerence MMM 2007). [14] L Y., A HZ, Huang C., Lao SH, Robust Head Trackng wth Partcles Based on Multle Cues Fuson, T.S. Huang et al. Eds.): HCI/ECCV 2006, LNCS 3979,.29-39, Srnger-Verlag Berln Hedelberg 2006.

Region Segmentation Readings: Chapter 10: 10.1 Additional Materials Provided

Region Segmentation Readings: Chapter 10: 10.1 Additional Materials Provided Regon Segmentaton Readngs: hater 10: 10.1 Addtonal Materals Provded K-means lusterng tet EM lusterng aer Grah Parttonng tet Mean-Shft lusterng aer 1 Image Segmentaton Image segmentaton s the oeraton of

More information

THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY

THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY Proceedngs of the 20 Internatonal Conference on Machne Learnng and Cybernetcs, Guln, 0-3 July, 20 THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY JUN-HAI ZHAI, NA LI, MENG-YAO

More information

Motivation. Motivation. Monte Carlo. Example: Soft Shadows. Outline. Monte Carlo Algorithms. Advanced Computer Graphics (Fall 2009)

Motivation. Motivation. Monte Carlo. Example: Soft Shadows. Outline. Monte Carlo Algorithms. Advanced Computer Graphics (Fall 2009) Advanced Comuter Grahcs Fall 29 CS 294, Renderng Lecture 4: Monte Carlo Integraton Rav Ramamoorth htt://nst.eecs.berkeley.edu/~cs294-3/a9 Motvaton Renderng = ntegraton Relectance equaton: Integrate over

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

Semi-Supervised Biased Maximum Margin Analysis for Interactive Image Retrieval

Semi-Supervised Biased Maximum Margin Analysis for Interactive Image Retrieval IP-06850-00.R3 Sem-Suervsed Based Maxmum Margn Analyss for Interactve Image Retreval Lnng Zhang,, Student Member, IEEE, Lo Wang, Senor Member, IEEE and Wes Ln 3, Senor Member, IEEE School of Electrcal

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

Multilayer Neural Networks and Nearest Neighbor Classifier Performances for Image Annotation

Multilayer Neural Networks and Nearest Neighbor Classifier Performances for Image Annotation (IJACSA) Internatonal Journal of Advanced Comuter Scence and Alcatons, Vol. 3, No. 11, 01 Multlayer Neural Networs and Nearest Neghbor Classfer erformances for Image Annotaton Mustaha OUJAOURA Laboratory

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

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

Hierarchical clustering for gene expression data analysis

Hierarchical clustering for gene expression data analysis Herarchcal clusterng for gene expresson data analyss Gorgo Valentn e-mal: valentn@ds.unm.t Clusterng of Mcroarray Data. Clusterng of gene expresson profles (rows) => dscovery of co-regulated and functonally

More information

Lecture notes: Histogram, convolution, smoothing

Lecture notes: Histogram, convolution, smoothing Lecture notes: Hstogram, convoluton, smoothng Hstogram. A plot o the ntensty dstrbuton n an mage. requency (# occurrences) ntensty The ollowng shows an example mage and ts hstogram: I we denote a greyscale

More information

Human Action Recognition Using Discriminative Models in the Learned Hierarchical Manifold Space

Human Action Recognition Using Discriminative Models in the Learned Hierarchical Manifold Space Human Acton Recognton Usng Dscrmnatve Models n the Learned Herarchcal Manfold Sace Le Han, We Lang*, nxao Wu and Yunde Ja School of Comuter Scence and Technology, Beng Insttute of Technology 5 South Zhongguancun

More information

Lecture Note 08 EECS 4101/5101 Instructor: Andy Mirzaian. All Nearest Neighbors: The Lifting Method

Lecture Note 08 EECS 4101/5101 Instructor: Andy Mirzaian. All Nearest Neighbors: The Lifting Method Lecture Note 08 EECS 4101/5101 Instructor: Andy Mrzaan Introducton All Nearest Neghbors: The Lftng Method Suose we are gven aset P ={ 1, 2,..., n }of n onts n the lane. The gven coordnates of the -th ont

More information

Recognition of Identifiers from Shipping Container Images Using Fuzzy Binarization and Enhanced Fuzzy Neural Network

Recognition of Identifiers from Shipping Container Images Using Fuzzy Binarization and Enhanced Fuzzy Neural Network Recognton of Identfers from Shng Contaner Images Usng uzzy Bnarzaton and Enhanced uzzy Neural Networ Kwang-Bae Km Det. of Comuter Engneerng, Slla Unversty, Korea gbm@slla.ac.r Abstract. In ths aer, we

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

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

An Improved Face Recognition Technique Based on Modular Multi-directional Two-dimensional Principle Component Analysis Approach

An Improved Face Recognition Technique Based on Modular Multi-directional Two-dimensional Principle Component Analysis Approach 301 JOURNAL OF SOFWARE, VOL. 9, NO. 1, DECEMBER 014 An Imroved Face Recognton echnque Based on Modular Mult-drectonal wo-dmensonal Prncle Comonent Analyss Aroach Deartment of Deartment of Xaoqng Dong Physcs

More information

Fuzzy C-Means Initialized by Fixed Threshold Clustering for Improving Image Retrieval

Fuzzy C-Means Initialized by Fixed Threshold Clustering for Improving Image Retrieval Fuzzy -Means Intalzed by Fxed Threshold lusterng for Improvng Image Retreval NAWARA HANSIRI, SIRIPORN SUPRATID,HOM KIMPAN 3 Faculty of Informaton Technology Rangst Unversty Muang-Ake, Paholyotn Road, Patumtan,

More information

NOVEL APPROACH FOR MOVING HUMAN DETECTION AND TRACKING IN STATIC CAMERA VIDEO SEQUENCES

NOVEL APPROACH FOR MOVING HUMAN DETECTION AND TRACKING IN STATIC CAMERA VIDEO SEQUENCES THE PUBLIHING HOUE PROCEEDING OF THE ROMANIAN ACADEMY, eres A, OF THE ROMANIAN ACADEMY Volume 3, Number 3/202,. 269 277 NOVEL APPROACH FOR MOVING HUMAN DETECTION AND TRACKING IN TATIC CAMERA VIDEO EQUENCE

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

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

A Method of Line Matching Based on Feature Points

A Method of Line Matching Based on Feature Points JOURNAL OF SOFTWARE, VOL. 7, NO. 7, JULY 2012 1539 A Method of Lne Matchng Based on Feature Ponts Yanxa Wang and Yan Ma College of Comuter and Informaton Scence, Chongqng Normal Unversty, Chongqng, 400047,

More information

FULL-FRAME VIDEO STABILIZATION WITH A POLYLINE-FITTED CAMCORDER PATH

FULL-FRAME VIDEO STABILIZATION WITH A POLYLINE-FITTED CAMCORDER PATH FULL-FRAME VIDEO STABILIZATION WITH A POLYLINE-FITTED CAMCORDER PATH Jong-Shan Ln ( 林蓉珊 ) We-Tng Huang ( 黃惟婷 ) 2 Bng-Yu Chen ( 陳炳宇 ) 3 Mng Ouhyoung ( 歐陽明 ) Natonal Tawan Unversty E-mal: {marukowetng}@cmlab.cse.ntu.edu.tw

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

Graph-based Clustering

Graph-based Clustering Graphbased Clusterng Transform the data nto a graph representaton ertces are the data ponts to be clustered Edges are eghted based on smlarty beteen data ponts Graph parttonng Þ Each connected component

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Why consder unlabeled samples?. Collectng and labelng large set of samples s costly Gettng recorded speech s free, labelng s tme consumng 2. Classfer could be desgned

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

Skew Estimation in Document Images Based on an Energy Minimization Framework

Skew Estimation in Document Images Based on an Energy Minimization Framework Skew Estmaton n Document Images Based on an Energy Mnmzaton Framework Youbao Tang 1, Xangqan u 1, e Bu 2, and Hongyang ang 3 1 School of Comuter Scence and Technology, Harbn Insttute of Technology, Harbn,

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

Optimized Query Planning of Continuous Aggregation Queries in Dynamic Data Dissemination Networks

Optimized Query Planning of Continuous Aggregation Queries in Dynamic Data Dissemination Networks WWW 007 / Trac: Performance and Scalablty Sesson: Scalable Systems for Dynamc Content Otmzed Query Plannng of Contnuous Aggregaton Queres n Dynamc Data Dssemnaton Networs Rajeev Guta IBM Inda Research

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

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

Machine Learning: Algorithms and Applications

Machine Learning: Algorithms and Applications 14/05/1 Machne Learnng: Algorthms and Applcatons Florano Zn Free Unversty of Bozen-Bolzano Faculty of Computer Scence Academc Year 011-01 Lecture 10: 14 May 01 Unsupervsed Learnng cont Sldes courtesy of

More information

On the Two-level Hybrid Clustering Algorithm

On the Two-level Hybrid Clustering Algorithm On the Two-level Clusterng Algorthm ng Yeow Cheu, Chee Keong Kwoh, Zongln Zhou Bonformatcs Research Centre, School of Comuter ngneerng, Nanyang Technologcal Unversty, Sngaore 639798 ezlzhou@ntu.edu.sg

More information

O n processors in CRCW PRAM

O n processors in CRCW PRAM PARALLEL COMPLEXITY OF SINGLE SOURCE SHORTEST PATH ALGORITHMS Mshra, P. K. Department o Appled Mathematcs Brla Insttute o Technology, Mesra Ranch-8355 (Inda) & Dept. o Electroncs & Electrcal Communcaton

More information

Application of Genetic Algorithms in Graph Theory and Optimization. Qiaoyan Yang, Qinghong Zeng

Application of Genetic Algorithms in Graph Theory and Optimization. Qiaoyan Yang, Qinghong Zeng 3rd Internatonal Conference on Materals Engneerng, Manufacturng Technology and Control (ICMEMTC 206) Alcaton of Genetc Algorthms n Grah Theory and Otmzaton Qaoyan Yang, Qnghong Zeng College of Mathematcs,

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

ACCURATE BIT ALLOCATION AND RATE CONTROL FOR DCT DOMAIN VIDEO TRANSCODING

ACCURATE BIT ALLOCATION AND RATE CONTROL FOR DCT DOMAIN VIDEO TRANSCODING ACCUATE BIT ALLOCATION AND ATE CONTOL FO DCT DOMAIN VIDEO TANSCODING Zhjun Le, Ncolas D. Georganas Multmeda Communcatons esearch Laboratory Unversty of Ottawa, Ottawa, Canada {lezj, georganas}@ mcrlab.uottawa.ca

More information

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning Outlne Artfcal Intellgence and ts applcatons Lecture 8 Unsupervsed Learnng Professor Danel Yeung danyeung@eee.org Dr. Patrck Chan patrckchan@eee.org South Chna Unversty of Technology, Chna Introducton

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

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

Image Segmentation. Image Segmentation

Image Segmentation. Image Segmentation Image Segmentaton REGION ORIENTED SEGMENTATION Let R reresent the entre mage regon. Segmentaton may be vewed as a rocess that arttons R nto n subregons, R, R,, Rn,such that n= R = R.e., the every xel must

More information

OUT-OF-SCENE AV DATA DETECTION

OUT-OF-SCENE AV DATA DETECTION OUT-OF-SCENE AV DATA DETECTION Danl Korchagn Ida Research Insttute P.O. Box 592, CH-1920 Martgny, Swtzerland ABSTRACT In ths aer, we roose a new aroach for the automatc audo-based out-of-scene detecton

More information

Tensor Locality Preserving Projections Based Urban Building Areas Extraction from High-Resolution SAR Images

Tensor Locality Preserving Projections Based Urban Building Areas Extraction from High-Resolution SAR Images Journal o Advances n Inormaton Technology Vol. 7, No. 4, November 016 Tensor Localty Preservng Proectons Based Urban Buldng Areas Extracton rom Hgh-Resoluton SAR Images Bo Cheng, Sha Cu, and Tng L Insttute

More information

A Vision-based Facial Expression Recognition and Adaptation System from Video Stream

A Vision-based Facial Expression Recognition and Adaptation System from Video Stream Internatonal Journal of achne Learnng and Comutng, Vol. 2, No. 5, October 202 A Vson-based Facal Exresson Recognton and Adataton System from Vdeo Stream Nurul Ahad awhd, Nasr Uddn Laskar, and Hader Al

More information

Machine Learning. Topic 6: Clustering

Machine Learning. Topic 6: Clustering Machne Learnng Topc 6: lusterng lusterng Groupng data nto (hopefully useful) sets. Thngs on the left Thngs on the rght Applcatons of lusterng Hypothess Generaton lusters mght suggest natural groups. Hypothess

More information

Bayesian Networks: Independencies and Inference. What Independencies does a Bayes Net Model?

Bayesian Networks: Independencies and Inference. What Independencies does a Bayes Net Model? Bayesan Networks: Indeendences and Inference Scott Daves and Andrew Moore Note to other teachers and users of these sldes. Andrew and Scott would be delghted f you found ths source materal useful n gvng

More information

Unsupervised Learning

Unsupervised Learning Pattern Recognton Lecture 8 Outlne Introducton Unsupervsed Learnng Parametrc VS Non-Parametrc Approach Mxture of Denstes Maxmum-Lkelhood Estmates Clusterng Prof. Danel Yeung School of Computer Scence and

More information

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work Outlne Seamless Image Sttchng n the Gradent Doman Anat Levn, Assaf Zomet, Shmuel Peleg and Yar Wess ECCV 004 Presenter: Pn Wu Oct 007 Introducton Related Work GIST: Gradent-doman Image Sttchng GIST GIST

More information

Solving Optimization Problems on Orthogonal Ray Graphs

Solving Optimization Problems on Orthogonal Ray Graphs Solvng Otmzaton Problems on Orthogonal Ray Grahs Steven Chalck 1, Phl Kndermann 2, Faban L 2, Alexander Wolff 2 1 Insttut für Mathematk, TU Berln, Germany chalck@math.tu-berln.de 2 Lehrstuhl für Informatk

More information

A High-Accuracy Algorithm for Surface Defect Detection of Steel Based on DAG-SVM

A High-Accuracy Algorithm for Surface Defect Detection of Steel Based on DAG-SVM Sensors & Transducers, Vol. 57, Issue 0, October 203,. 42-48 Sensors & Transducers 203 by IFSA htt://www.sensorsortal.com A Hgh-Accuracy Algorthm for Surface Defect Detecton of Steel Based on DAG-SVM,

More information

A new Algorithm for Lossless Compression applied to two-dimensional Static Images

A new Algorithm for Lossless Compression applied to two-dimensional Static Images A new Algorthm for Lossless Comresson aled to two-dmensonal Statc Images JUAN IGNACIO LARRAURI Deartment of Technology Industral Unversty of Deusto Avda. Unversdades, 4. 48007 Blbao SPAIN larrau@deusto.es

More information

Indirect Volume Rendering

Indirect Volume Rendering Indrect Volume Renderng Balázs Csébalv Deartment o Control Engneerng and Inormaton Technology Budaest Unversty o Technology and Economcs Classcaton o vsualzaton algorthms Drect Volume Renderng DVR: The

More information

An Entropy-Based Approach to Integrated Information Needs Assessment

An Entropy-Based Approach to Integrated Information Needs Assessment Dstrbuton Statement A: Approved for publc release; dstrbuton s unlmted. An Entropy-Based Approach to ntegrated nformaton Needs Assessment June 8, 2004 Wllam J. Farrell Lockheed Martn Advanced Technology

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

Design and Analysis of Algorithms

Design and Analysis of Algorithms Desgn and Analyss of Algorthms Heaps and Heapsort Reference: CLRS Chapter 6 Topcs: Heaps Heapsort Prorty queue Huo Hongwe Recap and overvew The story so far... Inserton sort runnng tme of Θ(n 2 ); sorts

More information

IMRT workflow. Optimization and Inverse planning. Intensity distribution IMRT IMRT. Dose optimization for IMRT. Bram van Asselen

IMRT workflow. Optimization and Inverse planning. Intensity distribution IMRT IMRT. Dose optimization for IMRT. Bram van Asselen IMRT workflow Otmzaton and Inverse lannng 69 Gy Bram van Asselen IMRT Intensty dstrbuton Webb 003: IMRT s the delvery of radaton to the atent va felds that have non-unform radaton fluence Purose: Fnd a

More information

A Clustering Algorithm for Key Frame Extraction Based on Density Peak

A Clustering Algorithm for Key Frame Extraction Based on Density Peak Journal of Computer and Communcatons, 2018, 6, 118-128 http://www.scrp.org/ournal/cc ISSN Onlne: 2327-5227 ISSN Prnt: 2327-5219 A Clusterng Algorthm for Key Frame Extracton Based on Densty Peak Hong Zhao

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

Contour Error of the 3-DoF Hydraulic Translational Parallel Manipulator. Ryszard Dindorf 1,a, Piotr Wos 2,b

Contour Error of the 3-DoF Hydraulic Translational Parallel Manipulator. Ryszard Dindorf 1,a, Piotr Wos 2,b Advanced Materals Research Vol. 874 (2014) 57-62 Onlne avalable snce 2014/Jan/08 at www.scentfc.net (2014) rans ech Publcatons, Swtzerland do:10.4028/www.scentfc.net/amr.874.57 Contour Error of the 3-DoF

More information

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints TPL-ware Dsplacement-drven Detaled Placement Refnement wth Colorng Constrants Tao Ln Iowa State Unversty tln@astate.edu Chrs Chu Iowa State Unversty cnchu@astate.edu BSTRCT To mnmze the effect of process

More information

Multi-view 3D Position Estimation of Sports Players

Multi-view 3D Position Estimation of Sports Players Mult-vew 3D Poston Estmaton of Sports Players Robbe Vos and Wlle Brnk Appled Mathematcs Department of Mathematcal Scences Unversty of Stellenbosch, South Afrca Emal: vosrobbe@gmal.com Abstract The problem

More information

LECTURE : MANIFOLD LEARNING

LECTURE : MANIFOLD LEARNING LECTURE : MANIFOLD LEARNING Rta Osadchy Some sldes are due to L.Saul, V. C. Raykar, N. Verma Topcs PCA MDS IsoMap LLE EgenMaps Done! Dmensonalty Reducton Data representaton Inputs are real-valued vectors

More information

Optimal Multiscale Organization of Multimedia Content for Fast Browsing and Cost-Effective Transmission

Optimal Multiscale Organization of Multimedia Content for Fast Browsing and Cost-Effective Transmission Proceedngs o the 6th WSEAS Internatonal Conerence on Sgnal Processng, Robotcs and Automaton, Coru Island, reece, February 6-9, 2007 257 Optmal Multscale Organzaton o Multmeda Content or Fast Browsng and

More information

S1 Note. Basis functions.

S1 Note. Basis functions. S1 Note. Bass functons. Contents Types of bass functons...1 The Fourer bass...2 B-splne bass...3 Power and type I error rates wth dfferent numbers of bass functons...4 Table S1. Smulaton results of type

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

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

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

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters Proper Choce of Data Used for the Estmaton of Datum Transformaton Parameters Hakan S. KUTOGLU, Turkey Key words: Coordnate systems; transformaton; estmaton, relablty. SUMMARY Advances n technologes and

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

Mircea cel Batran Naval Academy Scientific Bulletin, Volume XVI 2013 Issue 1 Published by Mircea cel Batran Naval Academy Press, Constanta, Romania

Mircea cel Batran Naval Academy Scientific Bulletin, Volume XVI 2013 Issue 1 Published by Mircea cel Batran Naval Academy Press, Constanta, Romania USING THE CHAOS THEORY AND DYNAMIC KEYS IN DIGITAL WATERMARKING Crstan-Gabrel APOSTOL 1 Dorn-Maran PÎRLOAGA Marus ROGOBETE 3 Cran RĂCUCIU 4 1 Eng. Ph.D., Mltary Electroncs and Inormatcs Systems Faculty,

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

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms Course Introducton Course Topcs Exams, abs, Proects A quc loo at a few algorthms 1 Advanced Data Structures and Algorthms Descrpton: We are gong to dscuss algorthm complexty analyss, algorthm desgn technques

More information

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization Problem efntons and Evaluaton Crtera for Computatonal Expensve Optmzaton B. Lu 1, Q. Chen and Q. Zhang 3, J. J. Lang 4, P. N. Suganthan, B. Y. Qu 6 1 epartment of Computng, Glyndwr Unversty, UK Faclty

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Supervsed vs. Unsupervsed Learnng Up to now we consdered supervsed learnng scenaro, where we are gven 1. samples 1,, n 2. class labels for all samples 1,, n Ths s also

More information

Iris Recognition Using Semantic Indexing

Iris Recognition Using Semantic Indexing Al-An et.al. Iraq Journal of Scence, December 0, Vol. 53, No. 4, P. 37-43 Irs Recognton Usng Semantc Indexng Lath A. Al-An *, Mohammed Sahb Altae, Ansam Ahmed Alwan Deartment of Physcs, College of Scence,

More information

Simulation Based Analysis of FAST TCP using OMNET++

Simulation Based Analysis of FAST TCP using OMNET++ Smulaton Based Analyss of FAST TCP usng OMNET++ Umar ul Hassan 04030038@lums.edu.pk Md Term Report CS678 Topcs n Internet Research Sprng, 2006 Introducton Internet traffc s doublng roughly every 3 months

More information

Computer models of motion: Iterative calculations

Computer models of motion: Iterative calculations Computer models o moton: Iteratve calculatons OBJECTIVES In ths actvty you wll learn how to: Create 3D box objects Update the poston o an object teratvely (repeatedly) to anmate ts moton Update the momentum

More information

STREET-SCENE TREE SEGMENTATION FROM MOBILE LASER SCANNING DATA

STREET-SCENE TREE SEGMENTATION FROM MOBILE LASER SCANNING DATA The Internatonal Archves of the Photogrammetry, Remote Sensng and Satal Informaton Scences, olume XLI-B3, 016 XXIII ISPRS Congress, 1 19 July 016, Prague, Czech Reublc STREET-SCENE TREE SEGMENTATION FROM

More information

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces Range mages For many structured lght scanners, the range data forms a hghly regular pattern known as a range mage. he samplng pattern s determned by the specfc scanner. Range mage regstraton 1 Examples

More information

Line Drawing Approach Based on Visual Curvature Estimation

Line Drawing Approach Based on Visual Curvature Estimation Lne Drawng Aroach Based on Vsual Curvature Estmaton Jun Lu, Mngquan Zhou, Guohua Geng, Feng Xao, Defa Hu Abstract In order to effectvely extract the feature lnes of the three-dmensonal model of the surface

More information

An Improved Image Segmentation Algorithm Based on the Otsu Method

An Improved Image Segmentation Algorithm Based on the Otsu Method 3th ACIS Internatonal Conference on Software Engneerng, Artfcal Intellgence, Networkng arallel/dstrbuted Computng An Improved Image Segmentaton Algorthm Based on the Otsu Method Mengxng Huang, enjao Yu,

More information

Broadcast Time Synchronization Algorithm for Wireless Sensor Networks Chaonong Xu 1)2)3), Lei Zhao 1)2), Yongjun Xu 1)2) and Xiaowei Li 1)2)

Broadcast Time Synchronization Algorithm for Wireless Sensor Networks Chaonong Xu 1)2)3), Lei Zhao 1)2), Yongjun Xu 1)2) and Xiaowei Li 1)2) Broadcast Tme Synchronzaton Algorthm for Wreless Sensor Networs Chaonong Xu )2)3), Le Zhao )2), Yongun Xu )2) and Xaowe L )2) ) Key Laboratory of Comuter Archtecture, Insttute of Comutng Technology Chnese

More information

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga Angle-Independent 3D Reconstructon J Zhang Mrelle Boutn Danel Alaga Goal: Structure from Moton To reconstruct the 3D geometry of a scene from a set of pctures (e.g. a move of the scene pont reconstructon

More information

A note on Schema Equivalence

A note on Schema Equivalence note on Schema Equvalence.H.M. ter Hofstede and H.. Proer and Th.P. van der Wede E.Proer@acm.org PUBLISHED S:.H.M. ter Hofstede, H.. Proer, and Th.P. van der Wede. Note on Schema Equvalence. Techncal Reort

More information

Index Terms-Software effort estimation, principle component analysis, datasets, neural networks, and radial basis functions.

Index Terms-Software effort estimation, principle component analysis, datasets, neural networks, and radial basis functions. ISO 9001:2008 Certfed Internatonal Journal of Engneerng and Innovatve Technology (IJEIT The Effect of Dmensonalty Reducton on the Performance of Software Cost Estmaton Models Ryadh A.K. Mehd College of

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

Potential Malicious Users Discrimination with Time Series Behavior Analysis

Potential Malicious Users Discrimination with Time Series Behavior Analysis Murat Semerc MURAT.SEMERCI@BOUN.EDU.TR Al Taylan Cemgl TAYLAN.CEMGIL@BOUN.EDU.TR Department of Computer Engneerng, Bogazc Unversty, 34342, Bebek, Istanbul, Turkey Bulent Sankur BULENT.SANKUR@BOUN.EDU.TR

More information

Biped Cartoon Retrieval Using LBG-Algorithm Based State Vector Quantization

Biped Cartoon Retrieval Using LBG-Algorithm Based State Vector Quantization Bed Cartoon Retreval Usng LBG-Algorthm Based State Vector Quantzaton Srraa Padee Deartment of Comuter Engneerng Facult of Engneerng Chulalongorn Unverst srraa490@ahoo.com Pzzanu Kanongchaos Deartment of

More information

Scale Selective Extended Local Binary Pattern For Texture Classification

Scale Selective Extended Local Binary Pattern For Texture Classification Scale Selectve Extended Local Bnary Pattern For Texture Classfcaton Yutng Hu, Zhlng Long, and Ghassan AlRegb Multmeda & Sensors Lab (MSL) Georga Insttute of Technology 03/09/017 Outlne Texture Representaton

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

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

Background Removal in Image indexing and Retrieval

Background Removal in Image indexing and Retrieval Background Removal n Image ndexng and Retreval Y Lu and Hong Guo Department of Electrcal and Computer Engneerng The Unversty of Mchgan-Dearborn Dearborn Mchgan 4818-1491, U.S.A. Voce: 313-593-508, Fax:

More information

Calibrating a single camera. Odilon Redon, Cyclops, 1914

Calibrating a single camera. Odilon Redon, Cyclops, 1914 Calbratng a sngle camera Odlon Redon, Cclops, 94 Our goal: Recover o 3D structure Recover o structure rom one mage s nherentl ambguous??? Sngle-vew ambgut Sngle-vew ambgut Rashad Alakbarov shadow sculptures

More information

Design of Structure Optimization with APDL

Design of Structure Optimization with APDL Desgn of Structure Optmzaton wth APDL Yanyun School of Cvl Engneerng and Archtecture, East Chna Jaotong Unversty Nanchang 330013 Chna Abstract In ths paper, the desgn process of structure optmzaton wth

More information

Transactions on Visualization and Computer Graphics. Sketching of Mirror-symmetric Shapes. Figure 1: Sketching of a symmetric shape.

Transactions on Visualization and Computer Graphics. Sketching of Mirror-symmetric Shapes. Figure 1: Sketching of a symmetric shape. Page of 0 Transactons on Vsualzaton and omuter Grahcs 0 0 0 Abstract Sketchng of Mrror-symmetrc Shaes For Peer Revew Only Ths aer resents a system to create mrror-symmetrc surfaces from sketches. The system

More information

Module Management Tool in Software Development Organizations

Module Management Tool in Software Development Organizations Journal of Computer Scence (5): 8-, 7 ISSN 59-66 7 Scence Publcatons Management Tool n Software Development Organzatons Ahmad A. Al-Rababah and Mohammad A. Al-Rababah Faculty of IT, Al-Ahlyyah Amman Unversty,

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

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 15

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 15 CS434a/541a: Pattern Recognton Prof. Olga Veksler Lecture 15 Today New Topc: Unsupervsed Learnng Supervsed vs. unsupervsed learnng Unsupervsed learnng Net Tme: parametrc unsupervsed learnng Today: nonparametrc

More information

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION Paulo Quntlano 1 & Antono Santa-Rosa 1 Federal Polce Department, Brasla, Brazl. E-mals: quntlano.pqs@dpf.gov.br and

More information