Laplacian Eigenmap for Image Retrieval

Size: px
Start display at page:

Download "Laplacian Eigenmap for Image Retrieval"

Transcription

1 Laplacan Egenmap for Image Retreval Xaofe He Partha Nyog Department of Computer Scence The Unversty of Chcago, 1100 E 58 th Street, Chcago, IL ABSTRACT Dmensonalty reducton has been receved much attenton n mage retreval. Even though the dmenson of mage feature vectors s normally very hgh, the embedded dmenson s much lower. Before we utlze any learnng technque, t s benefcal to frst perform dmensonalty reducton. Prncpal component analyss s one of the popular methods used, and can be shown to be optmal when the underlyng structure of data ponts s lnear. However, t fals to dscover the nonlnear structure f the data le on a lowdmensonal manfold embedded n hgh-dmensonal feature space. In ths paper, we apply a geometrcally motvated algorthm for representng the hgh dmensonal data, whch has localty preservng propertes. Consequently, a novel relevance feedback scheme on manfold s proposed. Experment results on real-world mage collectons have shown the effectveness and robustness of our proposed approaches. Keywords: mage retreval, laplacan egenmap, dmensonalty reducton, relevance feedback

2 1. Introducton Due to the rapd growth n the volume of dgt mages, there s an ncreasng demand for effectve mage management tools. Consequently, content-based mage retreval (CBIR) s recevng wdespread research nterest (Cox et al., 1996; Smth and Chang, 1996; Ma and Manjunath, 1997; Ru et al., 1997; Tong and Chang, 2001; Su et al, 2001). Content based mage retreval uses features automatcally extracted from the mages themselves, rather than manually provded annotatons, to facltate the retreval of mages relevant to a user s query. However, there are stll many open research ssues to be solved before such retreval system can be put nto practce. In recent years, much research has been done to deal wth the problem caused by the hgh dmensonalty of mage feature space (Tenenbaum et al., 2000; Rowes and Saul, 2000). Typcally, the dmensons of feature vector range from few tens to several hundreds. For example, a color hstogram may contan 256 bns. Hgh dmensonalty creates several problems for mage retreval. Frst, learnng from examples s computatonally nfeasble f t has to rely on hgh-dmensonal representatons. The reason for ths s known as curse of dmenson: the number of examples necessary for relable generalzaton grows exponentally wth the number of dmensons (Stone, 1982). Learnablty thus necesstates dmensonalty reducton. Second, n large multmeda databases, hgh-dmensonal representaton s computatonal ntensve and most users do not wat around to provde a great deal of feedbacks. Hence for storage and speed concern, dmensonalty reducton s needed. Two classcal technques for dmensonalty reducton are Prncple Component Analyss (PCA) and Multdmensonal Scalng (MDS). PCA performs dmensonalty reducton by projectng the orgnal n- dmensonal data onto the m(<n) dmensonal lnear subspace spanned by the leadng egenvectors of the data s covarance matrx. Thus PCA bulds a global lnear model of the data (an m dmensonal hyperplane).

3 For lnearly embedded manfolds, PCA s guaranteed to dscover the dmensonalty of the manfold and produce a compact representaton n the form of an orthonormal bass. However, PCA fals to dscover the underlyng structure, f the data le on a nonlnear submanfold of the feature space. For example, the 3 covarance matrx of data sampled from a helx n R wll have full-rank and thus three prncple components. However, the helx s a one-dmensonal manfold and can be parameterzed wth a sngle number. Classcal MDS fnds an embeddng that preserves parwse dstances between data ponts. It s equvalent to PCA when those dstances are Eucldean. A key queston n mage retreval s that, how do we judge smlarty? The choce of the smlarty measure for the nputs s a deep queston that les at the core of the feld of machne learnng. If the mages le on or close to a low-dmensonal nonlnear manfold embedded n hgh-dmensonal feature space, then the Eucldean dstance n hgh-dmensonal feature space may not accurately reflect ther ntrnsc smlarty, as measured by geodesc dstance along the low-dmensonal manfold. In addton, due to the gap between semantc concepts and low-level mage features, the global topology of mages can not match human percepton. Thus, local topology of mages s much more mportant and relable. When two mages are rrelevant, the absolute value of ther dstance n feature space gves us lttle nformaton about ther smlarty. For example, t s meanngless to say that, a tger s more lke a dog than a horse. In ths paper, we apply a smple nonlnear dmensonalty reducton algorthm, whch has localty preservng propertes. A lnear dmensonalty reducton algorthm s proposed as ts lnear extenson. Consequently, a geometrcally motvated relevance feedback scheme s proposed for mage rankng, whch s conducted on the mage manfold, rather than Eucldean space. To model the geodesc paths of the mage manfold, we fnd the shortest-path dstances between mages n the database, whch s n turn used as smlarty measure. Our goal s to dscover, gven only the unordered hgh-dmensonal nputs, low-dmensonal representatons that capture the ntrnsc degrees of freedom of an mage set. The rest of ths paper s organzed as follows. Secton 2 relates a lst of prevous works to our work. Secton 3 descrbes laplacan egenmap for nonlnear dmensonalty reducton. Secton 4 descrbes ts

4 connectons to prncple component analyss. In secton 5, we propose a lnear dmensonalty reducton algorthm as lnear extensons of laplacan egenmap. Secton 6 descrbes the proposed method for relevance feedback on manfold. The expermental results are shown n secton 7. Fnally, we gve conclusons and future work n secton Prevous Work One of the most popular models used n nformaton retreval s the vector space model (Salton and McGll, 1983). Varous retreval technques have been developed for ths model, ncludng the method of relevance feedback. Most prevous researches on relevance feedback have fallen nto the followng three categores: retreval based on query pont movement (Ru et al., 1997), retreval based on re-weghtng of dfferent feature dmenson (Ishkawa et al., 1998) and retreval based on updatng the probablty dstrbuton of mages n the database (Cox et al., 2000). In recent years, some learnng-based approaches are proposed. Wu et al. (2000) proposed a Dscrmnant-EM algorthm wthn the transductve learnng framework n whch both labeled and unlabeled mages are used. Teu et. al (2000) presented a framework for mage retreval based on representng mages wth a very large set of hghly selectve features. Queres are nteractvely learned onlne wth a smple boostng algorthm. Tong et. al (2001) proposed the use of a support vector machne actve learnng algorthm for conductng effectve relevance feedback for mage retreval. Whle most machne learnng algorthms are passve n the sense that they are generally appled usng a randomly selected tranng set, the SVM actve learnng algorthm chooses the most nformatve mages wthn the database to ask the user to label. All these approaches have acheved good emprcal results. However, a common lmtaton of them s that they do not consder the underlyng structure of mage set n hgh-dmensonal feature space. There have been several studes of dmensonalty reducton for mage retreval. In (Su et al., 2001), prncple component analyss s used to reduce both nose contaned n the orgnal mage features and dmenson of feature space. The PCA process s ncorporated nto the relevance feedback framework to

5 extract feature subspaces n order to represent the subjectve class mpled n the postve examples. Dfferent types of features (color, texture, etc.) are allowed to have dfferent dmensons accordng ther sgnfcances and dstrbutons as mpled n user s feedbacks. Cohen et al. (2002) proposes a novel method, called prncple feature analyss, for dmenson reducton of a feature set by choosng a subset of the orgnal features that contans most of the essental nformaton, usng the same crtera as the PCA. In (Wu, 2000), weghted mult-dmensonal scalng s used for dmensonalty reducton. All of these methods do not consder the structure of the mage manfold on whch the mages are lkely to resde. 3. Laplacan Egenmap for Image Representaton Recently, there has been some renewed nterest n the problem of developng low dmensonal representatons when data les on a manfold (Tenenbaum et al., 2000; Rowes and Saul, 2000). In mage retreval, t s desrable to map the mage feature vectors nto a reduced space, due to the consderaton of learnablty, memory storage, and computatonal speed. In addton, due to the gap between semantc concepts and low-level mage features, the global topology of mages can not always match human percepton. Thus, local topology of mages s much more mportant and relable. If the dstance of two mages s large enough, we regard them as rrelevant. When two mages are rrelevant, the absolute value of ther dstance gves us lttle nformaton about ther smlarty. For example, t s meanngless to say that, a tger s more lke a dog than a horse. Thus, for mage retreval, a crteron should be satsfed that, the mappng must have localty preservng propertes. That s, the nearest neghbors of an mage n the orgnal feature space should be mapped to nearest neghbors of that mage n the reduced space. Recall that prncpal component analyss (PCA) s a lnear mappng algorthm whch maxmzes the T varance. Let y = y, y,, } be a one-dmensonal map. The objectve functon of PCA s as follows { 1 2 L y n max y ( y y) 2 where y s the sample mean.

6 In ths paper, we apply laplacan egenmap (Belkn and Nyog, 2001) to map the mages nto a lowdmensonal space. Consder the problem of mappng the weghted graph G to a lne so that connected ponts stay as close as possble. Let y = y, y,, } be such a map. A reasonable crteron for choosng a { 1 2 L good map s to mnmze the followng objectve functon y n T j ( y y ) j 2 W j The mnmzaton problem reduces to fndng arg mn y T y Dy= 1 y T Ly where L = D W s the Laplacan matrx. D s dagonal weght matrx; ts entres are column (or row, snce W s symmetrc) sums of W, D = W j j. Laplacan s symmetrc, postve semdefnte matrx whch can be thought of as an operator on functons defned on vertces of G. W s the weght matrx such that W j 1 = 0 f ponts x, x j are connected otherwse The objectve functon wth our choce of weghts W j ncurs a heavy penalty f neghborng ponts x and x j are mapped far apart. Therefore, mnmzng t s an attempt to ensure that f x and x j are close then y and y j are close as well. The algorthmc procedure s formally stated below: Step 1 [Constructng the adjacency graph] Nodes and j are connected by an edge f s among k nearest neghbors of j or j s among k nearest neghbors of. Step 2 [Choosng the weghts] W f and only f vertces and j are connected by an edge. = 1 j

7 Step 3 [Egenmaps] Assume the graph G, constructed above, s connected, otherwse proceed wth Step 3 for each connected component. Compute egenvalues and egenvectors for the generalzed egenvector problem: Ly = λdy (1) Let y 0,..., y k 1 be the solutons of equaton 1, ordered accordng to ther egenvalues, Ly Ly Ly 0 1 LL = λ Dy 1 0 = λ Dy = λ k 1 k 1 k 1 0 = λ0 λ1 L λk Dy We leave out the egenvector y correspondng to egenvalue 0 and use the next m egenvectors for embeddng n m-dmensonal Eucldean space. 0 x ( y 1 ( ), L, y ( )) m 4. Connectons to Prncpal component analyss (PCA) 4.1 Prncpal component analyss Prncpal component analyss s a wdely used statstcal tool for data analyss. Gven a set of feature vectors n hgh-dmensonal space, the purpose s to fnd a low-dmensonal mappng n reduced space wth less redundancy, that would gve as good a representaton as possble. The redundancy s measured by correlatons between data elements. In the PCA transform, the feature vector x s frst centered by substractng ts mean: x x E(x)

8 The mean s estmated from the avalable samples x( 1), x(2), L, x( m) R n. Let us assume n the followng that the centerng has been done and thus E(x)=0. Let X denote the feature vector matrx, that s, the th row of X s the feature vector x(). So, X s a m n matrx. Thus, we can obtan the covarance matrx as follows: C = X T X By egenvector decomposton, C can be decomposed nto the product of three matrces: C = VΣV T where Σ = dag λ, λ, L, λ } are the egenvalues wth descendng order and V s a orthogonal matrx, whose { 1 2 n column vector s the correspondng egenvector of C. Thus, TPCA = V s the transformaton matrx of PCA. If we want to map these m-dmensonal feature vectors to a k-dmensonal (k<m) space, t could be done by smply set the m-k least egenvalues to be zero,.e., λ 0, L λ 0. k + 1, m By applyng Sngular Value Decomposton (SVD), we can reformulate ths problem as follows: T X = UΣV T where U and V are two orthogonal matrces, whose column vectors are the egenvector of XX and X T X (the covarance matrx C), respectvely, and Σ = dag λ, λ, L, λ } are the sngular values of X (also the { 1 2 n T egenvalues of XX and X T X ). Now, we obtan the transformed feature vector n new low-dmensonal space: X = = UΣ XT PCA Ths ndcates that the low-dmensonal representaton obtaned by PCA s just the product of two matrces: T (1) the matrx U whose column vectors are the leadng egenvectors of weght matrx W = XX : PCA W PCA y = λ y (2) PCA U = { y, y 2, L, y 1 k }

9 and (2) the dagonal matrx Σ whose dagonal elements are the leadng egenvalues of W, Σ = dag λ, λ, L, λ }. Note that the weght matrx W s measured by nner product of two mage PCA vectors. { 1 2 k PCA 4.2 Connectons to PCA In secton 4, we have show that the optmal embeddng for non-lnear case s obtaned by solvng a general egenvector problem below: Ly = λdy ( D W ) y = λdy Wy = (1 λ) Dy 1 D Wy = (1 λ) y 1 Let W D W, W s essentally a normalzed weght matrx preservng localty. (Note that the Laplacan = Laplacan matrx D provdes a natural measure on the vertces of the graph. The bgger the value D s, the more mportant s that vertex). We rewrte the above formula as follows: W Laplacan y = 1 λ ) y (3) ( Laplacan As we can see, the equatons (2) and (3) have the same form. Ths observaton shows that, wth the same weght matrx, Laplacan Egenmap and PCA wll yeld the same result, and λ PCA + λ Laplacan = 1. Therefore, the egenvector of W wth the largest egenvalue s just the egenvector of W wth the smallest egenvalue. PCA Laplacan The connectons also gve us a way to determne the dmensonalty of the low-dmensonal manfold, analogous to PCA. J ( k) = k (1 λ ) = 1 rank( W ) 1 = 1 (1 λ ) 100%

10 where J ( k) 90%. 0 = λ0 < λ1 λ2 L λ rank ( W ) 1. If we want to keep 90% nformaton, we smply choose such k that Based on above analyss, we conclude that, the essental dfference between laplacan egenmap and PCA s that they choose dfferent weght matrces. PCA uses nner product as a lnear smlarty measure, whle laplacan egenmap uses a non-lnear smlarty measure whch preserves localty. For PCA, ts advantage over laplacan egenmap s that t can produce a transform matrx T. Thus, for a new pont, t can be easly map to the new space. The dsadvantage s that, t fals to dscover the underlyng nonlnear structure of data set. PCA 5. Lnear Laplacan Egenmap As we descrbed prevously, one dsadvantage of laplacan egenmap s that t cannot produce a transformaton functon. That s, f the query mage s not n database, there s no way to map t nto the reduced space. To overcome ths problem, we propose lnear laplacan egenmap, whch s a lnear approxmaton of laplacan egenmap. Suppose A=(a 1,a 2,,a k ) s a transformaton matrx of lnear laplacan egenmap, that s, Y = XA, where the row vectors of matrx Y are the mage representatons n reduced space. We rewrte the mnmzaton problem as follows: arg mn ( XA) T L( XA) A T ( XA) D( XA) = I The vectors a that mnmze the above functon s gven by the mnmum egenvalue solutons to the generalzed egenvalue problem: L' a = λd'a where T L' = X LX, D' = X DX. T

11 Note that L and D are two n n matrces, where n s the dmensonalty of the orgnal space. Whle n the nonlnear case, L and D are two m m matrces, where m s the number of mages n database. In the real world applcatons, m >> n. Ths property makes lnear laplacan egenmap much more promsng than nonlnear laplacan egenmap. 6. Relevance Feedback on Image Manfold An mage retreval system should be able to rank the mages accordng to the query mage and relevance feedbacks. Tradtonally, the most wdely used dstance functon s Eucldean dstance. Ths s based on an assumpton that the mages le on a lnear hgh-dmensonal space. However, Eucldean dstance s not always a good choce, f the mages le on or close to a nonlnear manfold wth low-dmensonalty. Followng laplacan egenmap dscussed n secton 3 and lnear laplacan egenmap descrbed n secton 5, we propose a geometrcally motvated relevance feedback scheme for mage rankng, whch s conducted on manfold, rather than Eucldean space. Let R denote the set of query mage and relevant mages provded by the user. Our algorthm can be descrbed as follows: 1. Canddate generaton. For each mage x R, we fnd ts k-nearest neghbors C = { y1, y 2, L, y k } (those mages n R are excluded from selecton). Let C = C 1 U C 2 ULU C R. We call C canddate mage set. Note that R I C = φ. 2. Construct subgraph. We buld a subgraph H(V) wth weght matrx W, where V H = R U C.The dstance of any two vertces x, x j s measured as follows: dst( x, x j x ) = x j 2 f x s among x or x j s among x 's k' neareset neghbor otherwse j 's k' neareset neghbor,

12 Snce the mages n R are supposed to have some common semantcs, we set ther dstances to be zero. That s, dst(, x ) = 0, x, x R. Note that, k <k. x j j 3. Manfold dstance measure. To model the geodesc paths of the manfold, we fnd the shortestpath dstances between vertces n H. The length of a path n H s defned to be the sum of lnk weghts along that path. We then compute the geodesc dstance dst H ( x, x j ) (.e. shortest path length) between all pars of vertces of and j n H, usng Floyd s O ( r 3 ) algorthm. 4. Retreval based on geodesc dstance. To retreve the mages most smlar to the query, we smply sort them accordng to ther geodesc dstance to the query. The selected mages are presented to the user. 5. Update query example set. Add the relevant mages provded by the user nto R. Go back to step 1 untl the user s satsfed. From the above algorthm, we can see that the smlarty between the mage n database and the query mage s determned by ther geodesc dstance on mage manfold. The query examples (ncludng the orgnal query mage and relevant mages provded by the user) are merged nto one pont on the manfold. Thus, the geodesc dstance between query mage q and mage x can also be computed as follows: dst H ( q, x ) = mn{ dst ( r, x)} r R C where dst C ( r, x) s the length of the shortest path n C between mage r and mage x. That s, the vertces n the shortest path can only contan vertces n C. (Note that, dst H ( q, r) = 0, r R ) Thus, our algorthm s essentally dfferent from Roccho s relevance feedback scheme (1971). Roccho s formula for query refnement can be llustrated as follows: Q' = α Q 0 + β n R γ n 1 2 n S n = 1 1 = 1 2

13 where Q s the ntal query vector, Q ' s the refned query vector, R s the vector of relevant examples, 0 S s the vector of rrelevant examples, n and 1 n 2 are the number of relevant examples and rrelevant examples, respectvely; β and γ are sutable constants. Roccho s algorthm s essentally a perceptron-lke learnng algorthm when the nner product smlarty s used. Ths technque s mplemented n the MARS system (Ru et al., 1997). Roccho s smlarty-based relevance feedback algorthm has been wdely used n nformaton retreval, and has archved good emprcal results n text retreval doman. However, t may not be sutable for mage retreval, snce t fals to preserve localty. Due to the gap between semantc concepts and lowlevel mage features, the global topology of mages can not match human percepton. Thus, local dstrbuton of mages s much more mportant and relable. Hence the local nformaton should be preserved whle retreval. Fgure 1 llustrate a smple example: A stands for the ntal query mage; B s a relevant example provded by the user. C s the refned query by Roccho s algorthm ( α = β = 0. 5 ). Thus, mage D wll be retreved. However, by ntuton, mages E and F are preferred to mage D. In our proposed algorthm, A and B are merged nto a sngle pont. Thus, the geodesc dstance between A(B) and D s larger than the geodesc dstance between A(B) and E(F). So mage E and F wll be retreved, rather than mage D. 7. Expermental Results We performed several experments to evaluate the effectveness of the proposed approaches over a large mage database. The database we use conssts of 10,000 mages of 100 semantc categores from the Corel dataset. It s a large and heterogeneous mage set. A retreved mage s consdered correct f t belongs to the same category of the query mage. Three types of color features and three types of texture features are used n our system, whch are lsted n Table 1. The dmenson of the feature space s 435. We desgned an automatc feedback scheme to model the retreval process. At each teraton, the system makes the frst three ncorrect mages from the top 100 matches as rrelevant examples, and also selects at most 3 correct mages as relevant examples (relevant examples n the prevous teratons are

14 excluded from the selecton). These automatc generated feedbacks are added nto the query example set to refne the retreval. To evaluate the performance of our algorthms, we defne the retreval accuracy as follows: relevant mages retreved n top N returns Accuray = N 7.1 Three examples for 2-D data vsualzaton To compare the performance of laplacan egenmap, lnear laplacan egenmap and PCA, we show two examples. In the frst example, the dataset contans two classes. One class contans mage data of human eyes, and the other class contans mage data of non-human eyes. Each class contans 1500 mages n 400- dmensonal space. Each mage s an 8 bts (256 grey) mage of sze In the second example, a multclass dataset of handwrtten dgts ( 0-9 ) (Blake and Merz 1998). s used. The expermental results are shown n fgure Image Retreval n 2-dmensonal Reduced Space The mage database we use for ths experment conssts of 10,000 mages of 100 semantc categores from the Corel dataset. Each semantc category conssts of 100 mages. We compare the retreval result n PCA space, kernel PCA space, laplacan space and lnear laplacan space wth 2 dmensons. In PCA space, the tradtonal Roccho s relevance feedback scheme s used. In laplacan space and lnear laplacan space, our relevance feedback scheme on manfold s used. Fg 2 shows the experment results. As can be seen, our proposed relevance feedback scheme n laplacan space performs much better than Roccho s relevance feedback scheme n PCA space. 7.3 Image Retreval n Reduced Space wth dfferent dmensons In ths secton, we compared the performance of mage retreval n reduced space wth dfferent dmensons. Fgure 3 shows the experment results. As the feature space s drastcally reduced by PCA, the performance decreases fast. In PCA space wth 2 dmensons, only about 10% accuracy s acheved after

15 three teratons. Whle n 2-dmensonal laplacan space, we can stll acheve more than 40% accuracy after one teraton. Ths observaton shows that, our proposed algorthm s especally sutable for the case when drastc dmensonalty reducton needs to be performed. 8. Conclusons and Future Work Whle nonlnear dmensonalty reducton on manfold has receved much attenton, to the best of our knowledge we are not aware of any applcaton to mage representaton and retreval yet. In ths paper, we descrbed a nonlnear dmensonalty reducton algorthm and ts lnear extenson for mage retreval. The proposed scheme nfers a reduced space that preserves localty. A correspondng relevance feedback scheme s proposed. It s conducted on the mage manfold, rather than Eucldean space. As can be seen from the experments, our proposed algorthm performs much better than prncpal component analyss, especally when drastc dmensonalty reducton s performed. A fundamental problem n mage retreval s the gap between hgh-level semantc concepts and lowlevel mage features. Most exstng relevance feedback technques focus on mprovng the retreval performance of current query sesson, and the knowledge obtaned form the past user nteractons wth the system s forgotten. A possble extenson of our work s to somehow adjust the local topology of mages from user s relevance feedback, so the system wll gradually mprove ts retreval performance through accumulated user nteractons. We are currently explorng the effect and mpact of ths extenson.

16 References [1] Belkn, Mkhal and Nyog, Partha Laplacan egenmaps and spectral technques for embeddng and clusterng. NIPS, [2] Blake, C.L. and Merz, C.J. (1998). UCI repostory of machne learnng databases [ Irvne, CA: Unversty of Calforna, Department of Informaton and Computer Scence. [3] Cohen, I., Tan, Q., Zhou, X. S., and Huang, T. S Feature selecton usng prncpal feature analyss, submtted to ICIP 02. [4] Cox, I.J., Mnka, T.P., Papathomas, T.V., and Yanlos, P.N The Bayesan mage retreval system, PcHunter: theory, mplementaton, and psychophyscal experemtns, IEEE Trans. On Image Processng Specal Issue on Dgtal Lbrares, pp , vol. 9, No. 1. [5] Ishkawa, Y., Subramanya, R. and Faloutsos, C MndReader: query databases through multple examples. In proc. of the 24 th Int. Conf. on Very Large DataBases, pp [6] Ma, W.Y. and Manjunath, B.S Netra: A toolbox for navgatng large mage databases. In Proc. IEEE Internatonal Conference on Image Processng, Vol. 1, pp [7] Roccho, J. J., Relevance Feedback n Informaton Retreval. In the Smart System Experments n Automatc Document Processng, Englewood Clffs, NJ, Prentce Hall, pp [8] Rowes, S. T. and Sual, L. K Nonlnear Dmensonalty Reducton by Locally Lnear Embeddng, Scence, vol 290, 22. [9] Ru, Y., Huang, T.S., Mehrotra, S., and Ortega, M A relevance feedback archtecture for contentbased multmeda nformaton retreval systems. In Proc. IEEE workshop on content-based access of mage and vdeo lbrares, pp [10] Ru, Y., Huang, T. S., and Mehrotra, S Content-Based Image Retreval wth Relevance Feedback n MARS, n Proc. IEEE Int. Conf. on Image Processng.

17 [11] Salton, G. and McGll, M Introducton to modern nformaton retreval. McGraw-Hll, New York. [12] Smth, J.R. and Chang, S.-F., VsualSEEk: A fully automated Content-based Image Query System. ACM Multmeda Internatonal Conference, Boston, MA. [13] Stone, C. J Optmal global rates of convergence for nonparametrc regresson, Annals of statstcs, 10: [14] Su, Zhong, L, Stan, and Zhang, H Extracton feature subspaces for content-based retreval usng relevance feedback, ACM Multmeda. [15] Tenenbaum, J. B., Slva, V., and Langford, J. C A global geometrc framework for nonlnear dmensonalty reducton, Scence, vol 290, 22. [16] Teu, K. and Vola, P Boostng mage retreval, n Proc. IEEE Conf. CVPR, pp [17] Tong, S. and Chang, E Support vector machne actve learnng for mage retreval, n Proc. ACM Multmeda. [18] Wu. P., Manjunath., B.S., and Shn, H.D Dmensonalty Reducton for Image Retreval. In Proc. ICIP 00. [19] Wu, Y., Tan, Q., and Huang, T.S., Dscrmnant-EM algorthm wth applcaton to mage retreval, n Proc. IEEE Conf. CVPR.

18 Color-1 Color hstogram n HSV space wth quantzaton 256 Color-2 Frst and second moments n Lab space Color-3 Color coherence vector n LUV space wth quantzaton 64 Texture-1 Texture-2 Texture-3 Tamura coarseness hstogram Tamura drectonary Pyramd wavelet texture feature Table 1. Image features Fgure 1. A case where Roccho s relevance feedback gves a bad result. Image A s the ntal query example. Image B s a relevant example. Image C s the refned query. By Roccho s relevance feedback scheme, mage D wll be retreved, rather than E or F. Accuracy Retreval accuracy n reduced space wth 2 dmensons (top 10) number of rounds PCA space wth 2 dmensons Laplacan space wth 2 dmensons Orgnal space wth 435 dmensons Lnear Laplacan space wth 2 dmensons

19 Retreval accuracy n reduced space wth 2 dmensons (top 20) Accuracy PCA space wth 2 dmensons Laplacan space wth 2 dmensons Orgnal space wth 435 dmensons Lnear Laplacan space wth 2 dmensons number of rounds 6 Fgure 2. Image retreval n 2-dmensonal reduced space. Roccho s relevance feedback scheme s used n PCA space, whle our proposed algorthm s used n laplacan space, lnear laplacan space, and orgnal space. Accuracy Retreval accuracy n PCA space wth dfferent dmensons (top 10) 0 PCA 1 space wth 100 dmensons 2 PCA space wth 80 dmensons PCA space wth 40 dmensons PCA space wth 20 dmensons PCA space wth 10 dmensons PCA space wth 5 dmensons PCA space wth 2 dmensons number of rounds 3

20 Accuracy Retreval accuracy n Laplacan space wth dfferent dmensons (top 10) number of rounds Laplacan space wth 100 dmensons 3 Laplacan space wth 80 dmensons Laplacan space wth 40 dmensons Laplacan space wth 20 dmensons Laplacan space wth 10 dmensons Laplacan space wth 5 dmensons Laplacan space wth 2 dmensons Accuracy Retreval accuracy n Lnear Laplacan space wth dfferent dmensons (top 10) number of rounds 0 Lnear 1Laplacan space wth 100 2dmensons 3 Lnear Laplacan space wth 80 dmensons Lnear Laplacan space wth 40 dmensons Lnear Laplacan space wth 20 dmensons Lnear Laplacan space wth 10 dmensons Lnear Laplacan space wth 5 dmensons Lnear Laplacan space wth 2 dmensons Fgure 3. Image retreval n reduced space wth dfferent dmensons. Roccho s relevance feedback scheme s used n PCA space, whle our proposed algorthm s used n laplacan space and lnear laplacan space.

21

22 Fgure 4. Laplacan egenmap vs. PCA. (Two classes)

23 Fgure 5. Laplacan egenmap vs. PCA (10 classes, dgts 0-9 )

Learning an Image Manifold for Retrieval

Learning an Image Manifold for Retrieval Learnng an Image Manfold for Retreval Xaofe He*, We-Yng Ma, and Hong-Jang Zhang Mcrosoft Research Asa Bejng, Chna, 100080 {wyma,hjzhang}@mcrosoft.com *Department of Computer Scence, The Unversty of Chcago

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

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

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

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

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

A Unified Framework for Semantics and Feature Based Relevance Feedback in Image Retrieval Systems

A Unified Framework for Semantics and Feature Based Relevance Feedback in Image Retrieval Systems A Unfed Framework for Semantcs and Feature Based Relevance Feedback n Image Retreval Systems Ye Lu *, Chunhu Hu 2, Xngquan Zhu 3*, HongJang Zhang 2, Qang Yang * School of Computng Scence Smon Fraser Unversty

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

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

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION SHI-LIANG SUN, HONG-LEI SHI Department of Computer Scence and Technology, East Chna Normal Unversty 500 Dongchuan Road, Shangha 200241, P. R. Chna E-MAIL: slsun@cs.ecnu.edu.cn,

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

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

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

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

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

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

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

The Research of Support Vector Machine in Agricultural Data Classification

The Research of Support Vector Machine in Agricultural Data Classification The Research of Support Vector Machne n Agrcultural Data Classfcaton Le Sh, Qguo Duan, Xnmng Ma, Me Weng College of Informaton and Management Scence, HeNan Agrcultural Unversty, Zhengzhou 45000 Chna Zhengzhou

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

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

Recognizing Faces. Outline

Recognizing Faces. Outline Recognzng Faces Drk Colbry Outlne Introducton and Motvaton Defnng a feature vector Prncpal Component Analyss Lnear Dscrmnate Analyss !"" #$""% http://www.nfotech.oulu.f/annual/2004 + &'()*) '+)* 2 ! &

More information

Out-of-Sample Extensions for LLE, Isomap, MDS, Eigenmaps, and Spectral Clustering

Out-of-Sample Extensions for LLE, Isomap, MDS, Eigenmaps, and Spectral Clustering Out-of-Sample Extensons for LLE, Isomap, MDS, Egenmaps, and Spectral Clusterng Yoshua Bengo, Jean-Franços Paement, Pascal Vncent Olver Delalleau, Ncolas Le Roux and Mare Oumet Département d Informatque

More information

UB at GeoCLEF Department of Geography Abstract

UB at GeoCLEF Department of Geography   Abstract UB at GeoCLEF 2006 Mguel E. Ruz (1), Stuart Shapro (2), June Abbas (1), Slva B. Southwck (1) and Davd Mark (3) State Unversty of New York at Buffalo (1) Department of Lbrary and Informaton Studes (2) Department

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

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

Semi-Supervised Discriminant Analysis Based On Data Structure

Semi-Supervised Discriminant Analysis Based On Data Structure IOSR Journal of Computer Engneerng (IOSR-JCE) e-issn: 2278-0661,p-ISSN: 2278-8727, Volume 17, Issue 3, Ver. VII (May Jun. 2015), PP 39-46 www.osrournals.org Sem-Supervsed Dscrmnant Analyss Based On Data

More information

The Discriminate Analysis and Dimension Reduction Methods of High Dimension

The Discriminate Analysis and Dimension Reduction Methods of High Dimension Open Journal of Socal Scences, 015, 3, 7-13 Publshed Onlne March 015 n ScRes. http://www.scrp.org/journal/jss http://dx.do.org/10.436/jss.015.3300 The Dscrmnate Analyss and Dmenson Reducton Methods of

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

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

Learning a Locality Preserving Subspace for Visual Recognition

Learning a Locality Preserving Subspace for Visual Recognition Learnng a Localty Preservng Subspace for Vsual Recognton Xaofe He *, Shucheng Yan #, Yuxao Hu, and Hong-Jang Zhang Mcrosoft Research Asa, Bejng 100080, Chna * Department of Computer Scence, Unversty of

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

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

CS47300: Web Information Search and Management

CS47300: Web Information Search and Management CS47300: Web Informaton Search and Management Prof. Chrs Clfton 15 September 2017 Materal adapted from course created by Dr. Luo S, now leadng Albaba research group Retreval Models Informaton Need Representaton

More information

Orthogonal Complement Component Analysis for Positive Samples in SVM Based Relevance Feedback Image Retrieval

Orthogonal Complement Component Analysis for Positive Samples in SVM Based Relevance Feedback Image Retrieval Orthogonal Complement Component Analyss for ostve Samples n SVM Based Relevance Feedback Image Retreval Dacheng Tao and Xaoou Tang Department of Informaton Engneerng The Chnese Unversty of Hong Kong {dctao2,

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

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

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

Data-dependent Hashing Based on p-stable Distribution

Data-dependent Hashing Based on p-stable Distribution Data-depent Hashng Based on p-stable Dstrbuton Author Ba, Xao, Yang, Hachuan, Zhou, Jun, Ren, Peng, Cheng, Jan Publshed 24 Journal Ttle IEEE Transactons on Image Processng DOI https://do.org/.9/tip.24.2352458

More information

Relevance Feedback for Image Retrieval

Relevance Feedback for Image Retrieval Vashal D Dhale et al, / (IJCSIT Internatonal Journal of Computer Scence and Informaton Technologes, Vol 4 (2, 203, 39-323 Relevance Feedback for Image Retreval Vashal D Dhale, Dr A R Mahaan, Prof Uma Thakur

More information

GSLM Operations Research II Fall 13/14

GSLM Operations Research II Fall 13/14 GSLM 58 Operatons Research II Fall /4 6. Separable Programmng Consder a general NLP mn f(x) s.t. g j (x) b j j =. m. Defnton 6.. The NLP s a separable program f ts objectve functon and all constrants are

More information

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law)

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law) Machne Learnng Support Vector Machnes (contans materal adapted from talks by Constantn F. Alfers & Ioanns Tsamardnos, and Martn Law) Bryan Pardo, Machne Learnng: EECS 349 Fall 2014 Support Vector Machnes

More information

Semi-Supervised Kernel Mean Shift Clustering

Semi-Supervised Kernel Mean Shift Clustering IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. XX, NO. XX, JANUARY XXXX 1 Sem-Supervsed Kernel Mean Shft Clusterng Saket Anand, Student Member, IEEE, Sushl Mttal, Member, IEEE, Oncel

More information

Classification / Regression Support Vector Machines

Classification / Regression Support Vector Machines Classfcaton / Regresson Support Vector Machnes Jeff Howbert Introducton to Machne Learnng Wnter 04 Topcs SVM classfers for lnearly separable classes SVM classfers for non-lnearly separable classes SVM

More information

Manifold-Ranking Based Keyword Propagation for Image Retrieval *

Manifold-Ranking Based Keyword Propagation for Image Retrieval * Manfold-Rankng Based Keyword Propagaton for Image Retreval * Hanghang Tong,, Jngru He,, Mngjng L 2, We-Yng Ma 2, Hong-Jang Zhang 2 and Changshu Zhang 3,3 Department of Automaton, Tsnghua Unversty, Bejng

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

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

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

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

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

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

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

Learning-Based Top-N Selection Query Evaluation over Relational Databases

Learning-Based Top-N Selection Query Evaluation over Relational Databases Learnng-Based Top-N Selecton Query Evaluaton over Relatonal Databases Lang Zhu *, Wey Meng ** * School of Mathematcs and Computer Scence, Hebe Unversty, Baodng, Hebe 071002, Chna, zhu@mal.hbu.edu.cn **

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

MULTI-VIEW ANCHOR GRAPH HASHING

MULTI-VIEW ANCHOR GRAPH HASHING MULTI-VIEW ANCHOR GRAPH HASHING Saehoon Km 1 and Seungjn Cho 1,2 1 Department of Computer Scence and Engneerng, POSTECH, Korea 2 Dvson of IT Convergence Engneerng, POSTECH, Korea {kshkawa, seungjn}@postech.ac.kr

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

Query Clustering Using a Hybrid Query Similarity Measure

Query Clustering Using a Hybrid Query Similarity Measure Query clusterng usng a hybrd query smlarty measure Fu. L., Goh, D.H., & Foo, S. (2004). WSEAS Transacton on Computers, 3(3), 700-705. Query Clusterng Usng a Hybrd Query Smlarty Measure Ln Fu, Don Hoe-Lan

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

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

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

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

High-Boost Mesh Filtering for 3-D Shape Enhancement

High-Boost Mesh Filtering for 3-D Shape Enhancement Hgh-Boost Mesh Flterng for 3-D Shape Enhancement Hrokazu Yagou Λ Alexander Belyaev y Damng We z Λ y z ; ; Shape Modelng Laboratory, Unversty of Azu, Azu-Wakamatsu 965-8580 Japan y Computer Graphcs Group,

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

NAG Fortran Library Chapter Introduction. G10 Smoothing in Statistics

NAG Fortran Library Chapter Introduction. G10 Smoothing in Statistics Introducton G10 NAG Fortran Lbrary Chapter Introducton G10 Smoothng n Statstcs Contents 1 Scope of the Chapter... 2 2 Background to the Problems... 2 2.1 Smoothng Methods... 2 2.2 Smoothng Splnes and Regresson

More information

SELECTION OF THE NUMBER OF NEIGHBOURS OF EACH DATA POINT FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM

SELECTION OF THE NUMBER OF NEIGHBOURS OF EACH DATA POINT FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM ISSN 392 24X INFORMATION TECHNOLOGY AND CONTROL, 2007, Vol.36, No.4 SELECTION OF THE NUMBER OF NEIGHBOURS OF EACH DATA POINT FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM Rasa Karbauskatė,2, Olga Kurasova,2,

More information

An Improved Spectral Clustering Algorithm Based on Local Neighbors in Kernel Space 1

An Improved Spectral Clustering Algorithm Based on Local Neighbors in Kernel Space 1 DOI: 10.98/CSIS110415064L An Improved Spectral Clusterng Algorthm Based on Local Neghbors n Kernel Space 1 Xnyue Lu 1,, Xng Yong and Hongfe Ln 1 1 School of Computer Scence and Technology, Dalan Unversty

More information

Infrared face recognition using texture descriptors

Infrared face recognition using texture descriptors Infrared face recognton usng texture descrptors Moulay A. Akhlouf*, Abdelhakm Bendada Computer Vson and Systems Laboratory, Laval Unversty, Quebec, QC, Canada G1V0A6 ABSTRACT Face recognton s an area of

More information

Classification of Face Images Based on Gender using Dimensionality Reduction Techniques and SVM

Classification of Face Images Based on Gender using Dimensionality Reduction Techniques and SVM Classfcaton of Face Images Based on Gender usng Dmensonalty Reducton Technques and SVM Fahm Mannan 260 266 294 School of Computer Scence McGll Unversty Abstract Ths report presents gender classfcaton based

More information

Private Information Retrieval (PIR)

Private Information Retrieval (PIR) 2 Levente Buttyán Problem formulaton Alce wants to obtan nformaton from a database, but she does not want the database to learn whch nformaton she wanted e.g., Alce s an nvestor queryng a stock-market

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

Performance Evaluation of Information Retrieval Systems

Performance Evaluation of Information Retrieval Systems Why System Evaluaton? Performance Evaluaton of Informaton Retreval Systems Many sldes n ths secton are adapted from Prof. Joydeep Ghosh (UT ECE) who n turn adapted them from Prof. Dk Lee (Unv. of Scence

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

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

Fuzzy Filtering Algorithms for Image Processing: Performance Evaluation of Various Approaches

Fuzzy Filtering Algorithms for Image Processing: Performance Evaluation of Various Approaches Proceedngs of the Internatonal Conference on Cognton and Recognton Fuzzy Flterng Algorthms for Image Processng: Performance Evaluaton of Varous Approaches Rajoo Pandey and Umesh Ghanekar Department of

More information

Supervised Nonlinear Dimensionality Reduction for Visualization and Classification

Supervised Nonlinear Dimensionality Reduction for Visualization and Classification IEEE Transactons on Systems, Man, and Cybernetcs Part B: Cybernetcs 1 Supervsed Nonlnear Dmensonalty Reducton for Vsualzaton and Classfcaton Xn Geng, De-Chuan Zhan, and Zh-Hua Zhou, Member, IEEE Abstract

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

Feature Extraction and Selection for Image Retrieval

Feature Extraction and Selection for Image Retrieval Feature Extracton and Selecton for Image Retreval Xang Sean Zhou, Ira Cohen, Q Tan, Thomas S. Huang Beckman Insttute for Advanced Scence and Technology Unversty of Illnos at Urbana Champagn Urbana, IL

More information

Flatten a Curved Space by Kernel: From Einstein to Euclid

Flatten a Curved Space by Kernel: From Einstein to Euclid Flatten a Curved Space by Kernel: From Ensten to Eucld Quyuan Huang, Dapeng Olver Wu Ensten s general theory of relatvty fundamentally changed our vew about the physcal world. Dfferent from Newton s theory,

More information

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY SSDH: Sem-supervsed Deep Hashng for Large Scale Image Retreval Jan Zhang, and Yuxn Peng arxv:607.08477v2 [cs.cv] 8 Jun 207 Abstract Hashng

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

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices Internatonal Mathematcal Forum, Vol 7, 2012, no 52, 2549-2554 An Applcaton of the Dulmage-Mendelsohn Decomposton to Sparse Null Space Bases of Full Row Rank Matrces Mostafa Khorramzadeh Department of Mathematcal

More information

Structure from Motion

Structure from Motion Structure from Moton Structure from Moton For now, statc scene and movng camera Equvalentl, rgdl movng scene and statc camera Lmtng case of stereo wth man cameras Lmtng case of multvew camera calbraton

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

Lobachevsky State University of Nizhni Novgorod. Polyhedron. Quick Start Guide

Lobachevsky State University of Nizhni Novgorod. Polyhedron. Quick Start Guide Lobachevsky State Unversty of Nzhn Novgorod Polyhedron Quck Start Gude Nzhn Novgorod 2016 Contents Specfcaton of Polyhedron software... 3 Theoretcal background... 4 1. Interface of Polyhedron... 6 1.1.

More information

Optimal Workload-based Weighted Wavelet Synopses

Optimal Workload-based Weighted Wavelet Synopses Optmal Workload-based Weghted Wavelet Synopses Yoss Matas School of Computer Scence Tel Avv Unversty Tel Avv 69978, Israel matas@tau.ac.l Danel Urel School of Computer Scence Tel Avv Unversty Tel Avv 69978,

More information

Lecture 5: Multilayer Perceptrons

Lecture 5: Multilayer Perceptrons Lecture 5: Multlayer Perceptrons Roger Grosse 1 Introducton So far, we ve only talked about lnear models: lnear regresson and lnear bnary classfers. We noted that there are functons that can t be represented

More information

Two-Dimensional Supervised Discriminant Projection Method For Feature Extraction

Two-Dimensional Supervised Discriminant Projection Method For Feature Extraction Appl. Math. Inf. c. 6 No. pp. 8-85 (0) Appled Mathematcs & Informaton cences An Internatonal Journal @ 0 NP Natural cences Publshng Cor. wo-dmensonal upervsed Dscrmnant Proecton Method For Feature Extracton

More information

A Robust LS-SVM Regression

A Robust LS-SVM Regression PROCEEDIGS OF WORLD ACADEMY OF SCIECE, EGIEERIG AD ECHOLOGY VOLUME 7 AUGUS 5 ISS 37- A Robust LS-SVM Regresson József Valyon, and Gábor Horváth Abstract In comparson to the orgnal SVM, whch nvolves a quadratc

More information

Hermite Splines in Lie Groups as Products of Geodesics

Hermite Splines in Lie Groups as Products of Geodesics Hermte Splnes n Le Groups as Products of Geodescs Ethan Eade Updated May 28, 2017 1 Introducton 1.1 Goal Ths document defnes a curve n the Le group G parametrzed by tme and by structural parameters n the

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

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

Problem Set 3 Solutions

Problem Set 3 Solutions Introducton to Algorthms October 4, 2002 Massachusetts Insttute of Technology 6046J/18410J Professors Erk Demane and Shaf Goldwasser Handout 14 Problem Set 3 Solutons (Exercses were not to be turned n,

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

Incremental Learning with Support Vector Machines and Fuzzy Set Theory

Incremental Learning with Support Vector Machines and Fuzzy Set Theory The 25th Workshop on Combnatoral Mathematcs and Computaton Theory Incremental Learnng wth Support Vector Machnes and Fuzzy Set Theory Yu-Mng Chuang 1 and Cha-Hwa Ln 2* 1 Department of Computer Scence and

More information

Lecture #15 Lecture Notes

Lecture #15 Lecture Notes Lecture #15 Lecture Notes The ocean water column s very much a 3-D spatal entt and we need to represent that structure n an economcal way to deal wth t n calculatons. We wll dscuss one way to do so, emprcal

More information

Improving Web Image Search using Meta Re-rankers

Improving Web Image Search using Meta Re-rankers VOLUME-1, ISSUE-V (Aug-Sep 2013) IS NOW AVAILABLE AT: www.dcst.com Improvng Web Image Search usng Meta Re-rankers B.Kavtha 1, N. Suata 2 1 Department of Computer Scence and Engneerng, Chtanya Bharath Insttute

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

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

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS ARPN Journal of Engneerng and Appled Scences 006-017 Asan Research Publshng Network (ARPN). All rghts reserved. NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS Igor Grgoryev, Svetlana

More information

Load Balancing for Hex-Cell Interconnection Network

Load Balancing for Hex-Cell Interconnection Network Int. J. Communcatons, Network and System Scences,,, - Publshed Onlne Aprl n ScRes. http://www.scrp.org/journal/jcns http://dx.do.org/./jcns.. Load Balancng for Hex-Cell Interconnecton Network Saher Manaseer,

More information