What Is the Most Efficient Way to Select Nearest Neighbor Candidates for Fast Approximate Nearest Neighbor Search?

Size: px
Start display at page:

Download "What Is the Most Efficient Way to Select Nearest Neighbor Candidates for Fast Approximate Nearest Neighbor Search?"

Transcription

1 IEEE Internatonal Conference on Computer Vson What Is the Most Effcent Way to Select Nearest Neghbor Canddates for Fast Approxmate Nearest Neghbor Search? Masakazu Iwamura, Tomokazu Sato and Koch Kse Graduate School of Engneerng, Osaka Prefecture Unversty {masa, Abstract Approxmate nearest neghbor search (ANNS) s a basc and mportant technque used n many tasks such as object recognton. It nvolves two processes: selectng nearest neghbor canddates and performng a brute-force search of these canddates. Only the former though has scope for mprovement. In most exstng methods, t approxmates the space by quantzaton. It then calculates all the dstances between the query and all the quantzed values (e.g., clusters or bt sequences), and selects a fxed number of canddates close to the query. The performance of the method s evaluated based on accuracy as a functon of the number of canddates. Ths evaluaton seems ratonal but poses a serous problem; t gnores the computatonal cost of the process of selecton. In ths paper, we propose a new ANNS method that takes nto account costs n the selecton process. Whereas exstng methods employ computatonally expensve technques such as comparatve sort and heap, the proposed method does not. Ths realzes a sgnfcantly more effcent search. We have succeeded n reducng computaton tmes by one-thrd compared wth the state-of-theart on an experment usng mllon SIFT features.. Introducton Fndng the nearest neghbor (NN) of a gven query, called nearest neghbor search (NNS), s a smple but mportant task. An approxmate nearest neghbor search (ANNS) s an NNS n whch an approxmaton s ntroduced. ANNS problems are actvely studed and wdely appled to many applcatons such as near-duplcate detecton [], largescale object recognton [,,,, ], document mage retreval [], camera-based character recognton [, ]. The requred features for ANNS methods are accuracy n fndng the true NNs, and effcency n tme and space. In ths paper, we focus on the relatonshp between accuracy and computatonal effcency. Takng nto account the relatonshp between accuracy and computaton tmes s nherent n NNS problems. If the computaton tme s not an ssue, NNS problem can always be solved by a brute-force search,.e., calculatng all the dstances between the query and the data. Ths nave soluton s not practcal because t can take a long tme, especally for large datasets. A way to reduce computaton tmes s to reduce the data submtted to the brute-force search. That s, a lmted number of data, where we call NN canddates, s selected n the selecton process followed by the brute-force search. Only n the selecton process s there occason to ntroduce mprovements n the NNS problem and t needs to be fast and accurate. Lengthy computaton tmes n selectng the NN canddates negates the advantage n reducng the data. By not beng accurate n selectng NN canddates, the number of NN canddates needs augmentng to compensate for the low accuracy. Computaton tmes though would lengthen n the brute-force searches. The performance of most exstng methods s evaluated based on accuracy as a functon of the number of canddates. The number of canddates determnes computaton tmes for brute-force searches. However, t s ndependent of the computaton tmes of the selecton process. Hence papers gnorng the computaton tme of the selecton process tackle only a part of the ANNS problem. Some papers take nto account computaton cost properly (e.g., [,, ]). Among them, the most effcent method s the nverted mult-ndex () []. However, on the queston, What s the most effcent way to select NN canddates?, we found that the method does not capture the essence of the NN canddate selecton. An ntutve noton of what we mean s the followng. Imagne that there are n data and a task to fnd NN canddates of sze k (k <n). The worst approach s sortng them all, a task whch takes tme O(n log n). A better way s to use a partal sort technque usng a prorty queue. That s, each datum s added to the prorty queue and the best k data are kept. Ths s One mght beleve that computaton tmes for the selecton process can be gnored n comparson wth that for the brute-force search. However, a good ANNS method reduces the computaton tmes for brute-force searches. As a result, these computaton tmes are not the domnant factor. -/ $. IEEE DOI./ICCV..

2 known to take tme O(n log k). takes ths approach. However, there s a faster sort algorthm the bucket sort. It can complete the task n tme O(n) because t does not perform comparsons between data. Thus f we can prepare approprate buckets, the selecton of NN canddates can be much faster. Ths makes sense because our purpose s just fndng NN canddates; there s no need to arrange buckets n the ascendng order of ther dstances. Towards a practcally fast ANNS method for large scale datasets, we propose a more effcent ANNS method compared wth others, whch nvolves the most effcent way of fndng NN canddates. The proposed method s named bucket dstance hashng (). In the experments, we compared the proposed method wth varous representatve ANNS methods on the same platform wth respect to the crteron, recall as a functon of computaton tme, n addton to the commonly used crteron, recall as a functon of the number of canddates. Although we focus on the NN search, the same technque s drectly applcable to the k- NN search problem.. Related Work For an effcent selecton of NN canddates, data are usually ndexed n advance. Based on the structure for ndexng, ANN methods can be dvded nto tree-based and hashbased. Hash-based methods are also classed ether as data ndependent or data dependent. The former does not use data for ndexng though the latter does... Tree-based approach FLANN s a representatve tree-based method []. It automatcally selects the best method among the randomzed kd-tree [], the herarchcal k-means [] and the brute-force search, and tunes parameters for a gven dataset. In the experments, randomzed kd-tree and herarchcal k-means were compared wth the proposed method as FLANN tself dd not work for large datasets... Hash-based data ndependent approach Localty senstve hashng (LSH) s a representatve hash-based data ndependent method [, ]. There are many mprovements (e.g., [, ]). However, ts performance s known to be worse than the data dependent ndexng methods. We also confrmed that LSH s far slower than the data dependent methods that follow... Hash-based data dependent approach The data dependent methods can be categorzed nto those performng the selecton of NN canddates n the Eucldean space and the Hammng space. Selecton n the Eucldean space In ths category, the methods use quantzaton and data compresson technques. Vector quantzaton (VQ) s used n some ANNS methods such as VQ-ndex [7] and []. Product quantzaton (PQ) s used n []. As mentoned above, outperforms. Transform codng s used n []. The process can be regarded as equvalent to scalar quantzaton (SQ). From the category, we selected and for the evaluaton; both are revewed n the next secton. Although we mplemented transform codng, t was not comparable to those two. Selecton n the Hammng space In ths category, data are represented by bnary codes. Ther use helps reduce memory usage and computaton tmes n Hammngdstance calculatons because of the effcent btwse XOR operaton. Because of these good propertes, many ANNS methods based on bnary codes have been proposed (e.g., [,, 7,,, 7,, ]). Among them, spectral hashng (SH) [] s a representatve method. To acheve hgh recall, long codes, of length or bts long, are often used. Actually handlng such a long code s not easy. A baselne method for fndng NN canddates n Hammng space s the lnear search whch, whle fast, s ntractable when appled to large datasets. One mght thnk that the tree or hash structure helps to acheve sublnear searches. However, the number of data havng the same Hammng dstance to the query explode as the dstance ncreases, whch prevents the NN canddate selecton process from beng effcent []. As ponted out n [], one dea s to use the orgnal LSH for bts []. However, the data ndependent approach s neffcent as mentoned above. Another dea mght be to apply more sophstcated approaches used n exstng ANNS methods desgned for Eucldean space. We appled the proposed selecton method of NN canddates to SH. However, the performance dd not compare well wth other representatve methods.. Exstng Hash-based Indexng Methods Usng Vector and Product Quantzaton Revewng and n ths order helps understand the proposed method. Thus n ths secton, we present them n greater detal than Sec.... [] ndexes data usng VQ n selectng NN canddates effcently. Data are dvded nto clusters usng the k-means clusterng algorthm. Gven a query, the clusters close to the query are searched and data belongng to the clusters are selected as NN canddates. There s a possblty that some data close to the query are not ncluded n the NN canddates. However, VQ s known to be the best quantzaton method n respect to quantzaton error for the

3 same number of clusters []. Thus, t s expected to maxmze the accuracy of NN canddate selecton... Inverted mult-ndex () As a better soluton than, Babenko and Lemptsky proposed [], whch uses PQ nstead of VQ. PQ s an ntermedate quantzaton method between SQ and VQ; a vector space s dvded nto subspaces and then VQ s appled to the subvectors n each subspace. PQ generally produces larger errors than VQ wth the same number of clusters. Nevertheless, computaton tmes to acheve the same recall can be reduced wth the mult-sequence algorthm (MSA). Fgure llustrates an overvew of. Referrng to the fgure, there are four clusters n the subspace and three clusters n the subspace. The centrods of the clusters are represented by Cj, where labels each subspace and j labels a cluster n the -th subspace. Intally, the squared dstances between the query denoted by a star and the centrods n the subspaces are calculated. The squared dstance n the orgnal space s gven by the sum of the squared dstances n the subspaces. Then, f all the dstances n the orgnal space are calculated, the centrods close to the query n the orgnal space are found. However, t s not necessary to calculate all the dstances n the orgnal space, because centrods n subspaces wth large dstances cannot contrbute to the process. Hence, they are gnored n MSA. Fgure provdes an overvew of MSA. In MSA, the squared dstances n each subspace are sorted frst. Then, the centrods n the subspaces are examned n ascendng order of dstance. As shown n Fg. (a), the frst cluster examned s the drect product C C whose squared dstance s. The neghborng clusters C C and C C are selected as canddates to be examned n the next turn. Fgure (b) shows the next step when the canddate cluster C C whose squared dstance s s selected because ts squared dstance s the smallest among the canddates. Then, the cluster C C whose squared dstance s s selected as a new canddate. The reason that the cluster C C (wth squared dstance of but hdden n the fgure) s not selected as a canddate s that at least one canddate s guaranteed to have a smaller dstance than the cluster (see [] for more detal). Ths process lasts untl a suffcent number of NN canddates s obtaned. As seen above, MSA compares the dstances between clusters and arranges them n ascendng order of dstance. Ths s not requred n selectng the NN canddates. In ths regard, the method does not capture the essence of NN canddate selecton. The fgures and llustrate the case when the feature space s dvded nto two. If t s dvded nto more than two, the computatonal cost expands. Thus t s reported that performed best when the feature space s dvded nto two. We shall show that space dvson nto Squared dstances between the query and centrods n the orgnal subspace Centrods n subspace C C C C Squared dstances between the query and centrods n each subspace x Centrods n the orgnal space x C C C Centrods n subspace Query Fgure : Overvew of the nverted mult-ndex () when the feature space s dvded nto two. C C C C C C C C C C C C C C (a) C C C (d) C C C C C C C C (b) C C C (e) C C C C C C C C C C C (c) C C C Fgure : Overvew of the mult-sequence algorthm (MSA). The algorthm proceeds n the order from (a) to (f). The clusters n magenta represent those already selected and those n cyan are to be examned n the next turn. more than two acheves better performance wth a more effcent algorthm of selectng NN canddates.. Proposed Method.. Overvew As mentoned n Sec., the key dea of the proposed method s to select NN canddates wthout comparng data or clusters. We apply the branch and bound algorthm to the problem to whch MSA s appled. Fgure, correspondng to Fg., provdes an overvew of the proposed method. In the fgure, paths from query (Q) to the centrods (C) n the orgnal space are drawn. The left and rght halves of the paths represent the squared dstances {d j } between the query and centrods n subspaces and, respectvely. d j shares ts ndexes wth C j. Then, an upper bound of the squared dstance s determned. The (f) 7

4 C C C C C C C C C C C C C C C C C C C C C C C C C C C C Max 7 Max Max d = d d = d = d d = d = d = d = = d d = d = = d d d = = = d d = Q C Q C Q = = d d C Q d = = = d d C = = d = d = d = d = d = d = (a) (b) (c) (d) Fgure : Proposed algorthm when the upper bound of the squared dstance s. The algorthm proceeds n the order from (a) to (d). Query Q d = d = d = d = Squared dstances between the query and centrods n subspace d = d = d = Centrods n the orgnal space C Squared dstances between the query and centrods n subspace Fgure : Overvew of the proposed method when the feature space s dvded nto two. It corresponds to Fg.. Paths from query (Q) to centrods (C) n the orgnal space are drawn. upper bound s ncremented throughout the process. All the paths whose total dstances are less than an upper bound are selected n the followng process. Fgure llustrates the proposed algorthm when the upper bound of the squared dstance s. The algorthm proceeds n the order from (a) to (d). In (a), when the upper bound s set at, the path wth d = s mmedately removed because the path dstance s larger than the upper bound. Hereafter, each path n the subspace (left half) s examned. In (b), the path wth d =s examned. As the dfference between the path dstance and the upper bound s 7, the paths wth d =and d =on the subspace (rght half) are selected as NN canddates. In (c), the path wth d =s examned. As the dfference between the path dstance and the upper bound s, the paths wth d = and d =on the subspace (rght half) are selected as NN canddates. In (d), the path wth d =s examned. As the dfference between the path dstance and the upper bound s, the paths wth d =and d =on the subspace (rght half) are selected as NN canddates. Fnally, the same results are obtaned as n Fg.... Preparaton To explan the proposed method n more detal, some defntons and preparatons are gven. Let us assume both query, denoted by q, and data are represented as vectors. Let N and D be the number of data, and the dmensonalty of query and data, respectvely. We apply the prncpal component analyss (PCA) to the data so as to mnmze the estmaton error of dstance, and obtan the largest u prncpal components (egenvalues) l,l,...,l u n descendng order. Let V = [ v v... v u ] be the matrx of egenvectors correspondng to the u egenvalues. u corresponds to the dmensonalty of the mult-dmensonal hash table. We ntroduce a way to automatcally determne u later. Hereafter a cluster refers to a bucket... Indexng and parameter tunng PQ dvdes the feature space nto p-dmensonal subspaces,.e., a u-dmensonal vector x n the subspace spanned by u egenvectors s represented as x = {x,...,x m }, where m = u/p. Then, k-means clusterng s appled to each p-dmensonal subspace. Wth the excepton of p, we present an automatc parameter tunng algorthm. Let us begn wth the automatc parameter selecton for k-means clusterng. The purpose here s to dvde data nto k clusters n the -th subspace, and determne {k } and the centrods of the clusters n subspaces. Let x s be the -th subvector of the s-th datum, and C j be the j-th centrod n the -th subspace. Then, the quantza-

5 Algorthm : Adaptve quantzaton wth automatc parameter tunng. Ths algorthm fnds the number N bkt of buckets whch are closest to the number N of data. Input: p, D, N Output: N bkt, {k }, {Cj} m D/p ; k, for =,...,m; N bkt ; Calculate the ntal {Cj}, whch are means n subspaces; repeat // Preserve the current values of N bkt, {k } and {Cj} N bkt N bkt; {k } {k }; {Cj } {Cj}; // Increment by the number of clusters n the subspace havng the largest error Calculate {E }; t argmax E ; k t k t +; Update the centrods n the t-th subspace; // The number of buckets s determned by the product of the number of clusters k ; untl N bkt >N; f N/N bkt s closer to than N/N bkt then // Revert to the prevous values of N bkt, {k } and {Cj} N bkt N bkt; {k } {k }; {Cj} {C j }; end 7 N bkt m ton error E n the -th subspace s defned as N ( E = x s Cq (x s)), () s= where q (x) s the assgnment functon gven as ( ) q (x) =argmn x C j. () j Because a large quantzaton error causes a large estmaton error n dstance, we take up the strategy of mnmzng the largest quantzaton error E max n the subspaces. Algorthm outlnes the proposed algorthm. It ncreases the number of clusters n the subspace havng the largest quantzaton error, and ends when the total number of buckets approaches the number of data. As a result, the feature space s dvded nto N bkt buckets, whch corresponds to the hash sze of the mult-dmensonal hash table. We emprcally found the stoppng condton from experments over dfferent numbers of artfcal data followng the normal and unform dstrbutons and real ones such as SIFT and GIST features. The algorthm also determnes the dmensonalty u of the mult-dmensonal hash table. If the number k of clusters n a subspace s, the quantzaton error n the subspace s less than E max. In such a case, the subspace need no dvdng. Assumng the egenvalues are arranged n descendng order, there exsts m whch satsfes k > for =,...,m and k = for = m +,..., D/p. Then, gnorng the latter half ( m +), u s set to m p. Algorthm : Incremental search regon algorthm to effcently obtan the mnmum sze c of NN canddates. Input: {d j}, c Output: a lst of buckets contanng NN canddates n ; L ; U m b= mnj db j +Δ; // Pre-calculate the mnmum and maxmum partal bucket dstances n from the ( +)-thtothem-th subspaces for =to m do d mn m b=+ mnj db j; d max m b=+ maxj db j; end whle n<cdo // Fnd new NN canddates 7 n n+ CrcutBucketsFuncton(, ); // Extend the search regon L U; U U +Δ; end Algorthm : Functon CrcutBucketsFuncton(, d). Input: : ndex of subspace, d: partal bucket dstance Output: n: the number of NN canddates f <mthen // Fnd an ndex n the -th subspace satsfyng the condton, and then contnue the search n the ( +)-th subspace forall the j do f L d + d j + d max & d + d j + d mn <Uthen h j; CrcutBucketsFuncton( +,d+ d j); end 7 end else // Return the bucket ndexed by the m ndexes forall the j do f L d + d m j & d + d m j <Uthen Return the bucket whose hash value s H = {h,h,,h m} and the number of data n the bucket as n; end end end.. Effcent selecton of NN canddates Let c be the number of NN canddates requred. We propose an algorthm to select at least c NN canddates wth the smallest estmated dstances n a step-by-step manner. Algorthms and show the detaled procedure, a part of whch s already presented n Sec... Algorthm fnds buckets whose dstances are n the range between the lower bound L and upper bound U of the estmated dstance, callng recursvely Algorthm. The ntal values of L and U are and the squared dstance between the query and the nearest bucket, respectvely. As long as the number n of the

6 NN canddates obtaned so far s less than c, the process s repeated wth updated search regons. Δ s the ncrement for L and U. We used / of the sum of the egenvalues as Δ.. Experments We performed experments to compare the proposed method () wth representatve ANNS methods, whch nclude [] and [] as hash-based methods and randomzed kd-tree () [] and herarchcal k- means () [] as tree-based methods from the components of FLANN []. All the methods were mplemented n the C++ language. We mplemented the hash-based methods by ourselves, referrng to the MATLAB verson of and C++ source code of., and share most of the source code. Ths avods dfferences n mplementaton and makes for a farer comparson. For the tree-based methods, we used the C++ source code of FLANN obtaned at the authors homepage. FLANN tself dd not work for large datasets. We expermentally explored the best parameters for each method, and determned ones whch acheved the best performance n the crteron, recall as a functon of computaton tme. The parameters used n the experments are summarzed n Table. We used the BIGANN database contanng bllon -dmensonal SIFT descrptors [] and the Mllon Tny Images 7 contanng approxmately mllon - dmensonal GIST descrptors []. For the former, mllon, mllon and mllon datasets were used; data were used as queres. For the latter, thousand, mllon and mllon datasets were used; the frst data n the frst categores ( data n total) were removed from the datasets and used as queres. For both databases, the smallest dataset, whch s a subset of larger datasets, s used for tranng (parameter tunng). We employed servers where CPUs (AMD Opteron 7,.GHz, cores) and GB memory were nstalled. All data were stored on memory. Each program was executed as a sngle thread on a sngle core... Experment : Recall vs computaton tme We compared the methods n the crteron, recall as a functon of computaton tme. Computaton tme s defned as the average computaton tme requred to obtan an answer from the tme the query s gven. Accuraces for small Although LSH [], spectral hashng [], transform codng [] were examned, ther performance was too poor. Therefore, we have omtted them marusm/ndex.php/flann/flann computaton tmes were relatvely low because of lmtatons n measurng computaton tmes. Fgure shows expermental results on the SIFT and GIST datasets. They show that the proposed method performed much better than the others. Fgure (c) shows that was two tmes faster than and. tmes faster than n % recall, and. tmes faster than and. tmes faster than IV- FADC n % recall. Comparng Fgs. (a) to (c), the advantage of the proposed method became clearer as dataset sze ncreased. Ths shows scalablty of the proposed algorthm. In contrast, comparng Fgs. (d) to (f), the advantage of the proposed method does not change. Ths mght be due to the hgh dmensonalty of the feature vectors. The fact that and dd not scale s caused by relatvely large computatonal overhead. Lettng G be the number of clusters n the orgnal space, needs tme G for the dstance calculaton n selectng NN canddates. needs tme G for dstance calculaton, and tme G log G n sortng the dstances n the subspaces... Experment : Recall vs number of canddates We also compared the methods n the commonly used crteron, recall as a functon of the number of canddates. Wth ths crteron, we show the expermental results of hash-based methods. We selected the best parameters subject to the crteron, recall as a functon of computaton tme. Fgure shows that was also better than and IV- FADC n the standard crteron.. Conclusons In the approxmate nearest neghbor search (ANNS) process, only the selecton of nearest neghbor canddates has any leeway for mprovement. In ths paper, we ponted out the mportance of evaluatng the computatonal costs of ANNS methods. Because takng the computaton tme nto account s nherent n the ANNS problem, researches gnorng computaton tme have only a lmted mpact. We also ponted out that the state-of-the-art ANNS method,, does not capture the essence of NN canddate selecton. That s, although the goal of NN canddate selecton s smply fndng NN canddates, t employs unnecessary processes such as comparatve sort and heap. Fnally, we proposed a new ANNS method that takes nto account the cost of NN canddate selecton. The proposed method s based on the branch and bound algorthm whch produces a sgnfcantly more effcent search. We also presented an automatc parameter tunng algorthm whch, except for one parameter, can automatcally determne the best parameters. In an experment usng mllon SIFT features, the proposed method succeeded n reducng the computaton tme by one thrd compared wth the state-of-theart; the proposed method was two tmes faster n % recall, and. tmes faster n % recall.

7 Table : Parameters used n the experments. Methods Parameters SIFT M SIFT M SIFT M GIST K GIST M GIST M log C, P,,,,,, log C log C No. of trees k Not executable 7.. (a) SIFT M 7. (b) SIFT M 7. (c) SIFT M 7.. (d) GIST K 7. (e) GIST M 7. (f) GIST M Fgure : Experment : Recall vs Computaton tme. Acknowledgment Ths work was supported by JST CREST project, JST Funds for ntegrated promoton of socal system reform and research and development, JSPS KAKENHI #, and JST A-STEP #ASZH. References [] A. Andon and P. Indyk. Near-optmal hashng algorthms for approxmate nearest neghbor n hgh dmensons. In Proc. FOCS,. [] A. Babenko and V. Lemptsky. The nverted mult-ndex. In Proc. CVPR,.,,, [] J. Brandt. Transform codng for fast approxmate nearest neghbor search n hgh dmensons. In Proc. CVPR,., [] M. Datar, N. Immorlca, P. Indyk, and V. S. Mrrokn. Localty-senstve hashng scheme based on p-stable dstrbutons. In Proc. th annual symposum on Computatonal geometry,., [] A. Gersho and R. Gray. Vector Quantzaton and Sgnal Compresson. Kluwer Acadmc Publshers,. [] Y. Gong and S. Lazebnk. Iteratve quantzaton: A procrustean approach to learnng bnary codes. In Proc. CVPR,. [7] Y. Gong, S. Lazebnk, A. Gordo, and F. Perronnn. Iteratve quantzaton: A procrustean approach to learnng bnary codes for large-scale mage retreval. IEEE Trans. PAMI,. [] P. Indyk and R. Motwan. Approxmate nearest neghbor: towards removng the curse of dmensonalty. In Proc. th Symposum on Theory of Computng,. [] M. Iwamura, T. Kobayash, and K. Kse. Recognton of multple characters n a scene mage usng arrangement of local features. In Proc. ICDAR,. [] M. Iwamura, T. Tsuj, and K. Kse. Memory-based recognton of camera-captured characters. In Proc. DAS,.

8 7 (a) SIFT M 7 e+ (b) SIFT M 7 e+ (c) SIFT M 7 (d) GIST K 7 e+ (e) GIST M 7 e+ (f) GIST M Fgure : Experment : Recall vs Number of canddates. [] H. Jégou, M. Douze, and C. Schmd. Packng bag-offeatures. In Proc. ICCV,. [] H. Jégou, M. Douze, and C. Schmd. Product quantzaton for nearest neghbor search. IEEE Trans. PAMI, ():7,.,, [] Y. Ke, R. Sukthankar, and L. Huston. Effcent near-duplcate detecton and sub-mage retreval. In Proc. ACM Multmeda,. [] K. Kse, K. Noguch, and M. Iwamura. Robust and effcent recognton of low-qualty mages by cascaded recognzers wth massve local features. In Proc. WS-LAVD,. [] W. Kong and W.-J. L. Isotropc hashng. In Proc. NIPS,. [] B. Kuls and T. Darrell. Learnng to hash wth bnary reconstructve embeddngs. In Proc. NIPS,. [7] R.-S. Ln, D. A. Ross, and J. Yagnk. SPEC hashng: Smlarty preservng algorthm for entropy-based codng. In Proc. CVPR,. [] D. G. Lowe. Dstnctve mage features from scale-nvarant keyponts. IJCV, (),. [] Q. Lv, W. Josephson, Z. Wang, M. Charkar, and K. L. Mult-probe LSH: effcent ndexng for hgh-dmensonal smlarty search. In Proc. VLDB, 7. [] M. Muja and D. G. Lowe. Fast approxmate nearest neghbors wth automatc algorthm confguraton. In Proc. VISS- APP. INSTICC Press,.,, [] D. Nstér and H. Stewénus. Scalable recognton wth a vocabulary tree. In Proc. CVPR,.,, [] M. Norouz and D. J. Fleet. Mnmal loss hashng for compact bnary codes. In Proc. ICML,. [] A. Olva and A. Torralba. Modelng the shape of the scene: A holstc representaton of the spatal envelope. IJCV,,. [] C. Slpa-Anan and R. Hartley. Optmsed kd-trees for fast mage descrptor matchng. In Proc. CVPR,., [] K. Takeda, K. Kse, and M. Iwamura. Memory reducton for real-tme document mage retreval wth a mllon pages database. In Proc. CBDAR,. [] A. Torralba, R. Fergus, and W. T. Freeman. mllon tny mages: a large dataset for non-parametrc object and scene recognton. IEEE Trans. PAMI, (): 7,. [7] E. Tuncel, H. Ferhatosmanoglu, and K. Rose. VQ-ndex: an ndex structure for smlarty searchng n multmeda databases. In Proc. ACM Multmeda,. [] Y. Wess, R. Fergus, and A. Torralba. Multdmensonal spectral hashng. In Proc. ECCV, Part V (LNCS 77),. [] Y. Wess, A. Torralba, and R. Fergus. Spectral hashng. In Proc. NIPS,., [] H. Xu, J. Wang, Z. L, G. Zeng, S. L, and N. Yu. Complementary hashng for approxmate nearest neghbor search. In Proc. ICCV,. [] L. Zhang, Y. Zhang, J. Tang, K. Lu, and Q. Tan. Bnary code rankng wth weghted hammng dstance. In Proc. CVPR,.

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

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

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

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

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

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

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

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

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

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

Searching in one billion vectors: re-rank with source coding

Searching in one billion vectors: re-rank with source coding Searchng n one bllon vectors: re-rank wth source codng Hervé Jégou, Roman Tavenard, Matthjs Douze, Laurent Amsaleg To cte ths verson: Hervé Jégou, Roman Tavenard, Matthjs Douze, Laurent Amsaleg. Searchng

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. 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

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

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

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

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

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

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

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

Today s Outline. Sorting: The Big Picture. Why Sort? Selection Sort: Idea. Insertion Sort: Idea. Sorting Chapter 7 in Weiss.

Today s Outline. Sorting: The Big Picture. Why Sort? Selection Sort: Idea. Insertion Sort: Idea. Sorting Chapter 7 in Weiss. Today s Outlne Sortng Chapter 7 n Wess CSE 26 Data Structures Ruth Anderson Announcements Wrtten Homework #6 due Frday 2/26 at the begnnng of lecture Proect Code due Mon March 1 by 11pm Today s Topcs:

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

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

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

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

More information

Mathematics 256 a course in differential equations for engineering students

Mathematics 256 a course in differential equations for engineering students Mathematcs 56 a course n dfferental equatons for engneerng students Chapter 5. More effcent methods of numercal soluton Euler s method s qute neffcent. Because the error s essentally proportonal to the

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

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

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

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search Sequental search Buldng Java Programs Chapter 13 Searchng and Sortng sequental search: Locates a target value n an array/lst by examnng each element from start to fnsh. How many elements wll t need to

More information

Conditional Speculative Decimal Addition*

Conditional Speculative Decimal Addition* Condtonal Speculatve Decmal Addton Alvaro Vazquez and Elsardo Antelo Dep. of Electronc and Computer Engneerng Unv. of Santago de Compostela, Span Ths work was supported n part by Xunta de Galca under grant

More information

Hierarchical Image Retrieval by Multi-Feature Fusion

Hierarchical Image Retrieval by Multi-Feature Fusion Preprnts (www.preprnts.org) NOT PEER-REVIEWED Posted: 26 Aprl 207 do:0.20944/preprnts20704.074.v Artcle Herarchcal Image Retreval by Mult- Fuson Xaojun Lu, Jaojuan Wang,Yngq Hou, Me Yang, Q Wang* and Xangde

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

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

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

Sorting Review. Sorting. Comparison Sorting. CSE 680 Prof. Roger Crawfis. Assumptions

Sorting Review. Sorting. Comparison Sorting. CSE 680 Prof. Roger Crawfis. Assumptions Sortng Revew Introducton to Algorthms Qucksort CSE 680 Prof. Roger Crawfs Inserton Sort T(n) = Θ(n 2 ) In-place Merge Sort T(n) = Θ(n lg(n)) Not n-place Selecton Sort (from homework) T(n) = Θ(n 2 ) In-place

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

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

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

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

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

Programming in Fortran 90 : 2017/2018

Programming in Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Exercse 1 : Evaluaton of functon dependng on nput Wrte a program who evaluate the functon f (x,y) for any two user specfed values

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

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

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

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

More information

A Robust Method for Estimating the Fundamental Matrix

A Robust Method for Estimating the Fundamental Matrix Proc. VIIth Dgtal Image Computng: Technques and Applcatons, Sun C., Talbot H., Ourseln S. and Adraansen T. (Eds.), 0- Dec. 003, Sydney A Robust Method for Estmatng the Fundamental Matrx C.L. Feng and Y.S.

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

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

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

A Deflected Grid-based Algorithm for Clustering Analysis

A Deflected Grid-based Algorithm for Clustering Analysis A Deflected Grd-based Algorthm for Clusterng Analyss NANCY P. LIN, CHUNG-I CHANG, HAO-EN CHUEH, HUNG-JEN CHEN, WEI-HUA HAO Department of Computer Scence and Informaton Engneerng Tamkang Unversty 5 Yng-chuan

More information

K-means and Hierarchical Clustering

K-means and Hierarchical Clustering Note to other teachers and users of these sldes. Andrew would be delghted f you found ths source materal useful n gvng your own lectures. Feel free to use these sldes verbatm, or to modfy them to ft your

More information

Meta-heuristics for Multidimensional Knapsack Problems

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

More information

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

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

Linear Cross-Modal Hashing for Efficient Multimedia Search

Linear Cross-Modal Hashing for Efficient Multimedia Search Lnear Cross-Modal Hashng for Effcent Multmeda Search Xaofeng Zhu Z Huang Heng Tao Shen Xn Zhao College of CSIT, Guangx Normal Unversty, Guangx, 544,P.R.Chna School of ITEE, The Unversty of Queensland,

More information

An Efficient Genetic Algorithm with Fuzzy c-means Clustering for Traveling Salesman Problem

An Efficient Genetic Algorithm with Fuzzy c-means Clustering for Traveling Salesman Problem An Effcent Genetc Algorthm wth Fuzzy c-means Clusterng for Travelng Salesman Problem Jong-Won Yoon and Sung-Bae Cho Dept. of Computer Scence Yonse Unversty Seoul, Korea jwyoon@sclab.yonse.ac.r, sbcho@cs.yonse.ac.r

More information

Quality Improvement Algorithm for Tetrahedral Mesh Based on Optimal Delaunay Triangulation

Quality Improvement Algorithm for Tetrahedral Mesh Based on Optimal Delaunay Triangulation Intellgent Informaton Management, 013, 5, 191-195 Publshed Onlne November 013 (http://www.scrp.org/journal/m) http://dx.do.org/10.36/m.013.5601 Qualty Improvement Algorthm for Tetrahedral Mesh Based on

More information

Parallel matrix-vector multiplication

Parallel matrix-vector multiplication Appendx A Parallel matrx-vector multplcaton The reduced transton matrx of the three-dmensonal cage model for gel electrophoress, descrbed n secton 3.2, becomes excessvely large for polymer lengths more

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

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

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

More information

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

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE

SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE Dorna Purcaru Faculty of Automaton, Computers and Electroncs Unersty of Craoa 13 Al. I. Cuza Street, Craoa RO-1100 ROMANIA E-mal: dpurcaru@electroncs.uc.ro

More information

Sorting: The Big Picture. The steps of QuickSort. QuickSort Example. QuickSort Example. QuickSort Example. Recursive Quicksort

Sorting: The Big Picture. The steps of QuickSort. QuickSort Example. QuickSort Example. QuickSort Example. Recursive Quicksort Sortng: The Bg Pcture Gven n comparable elements n an array, sort them n an ncreasng (or decreasng) order. Smple algorthms: O(n ) Inserton sort Selecton sort Bubble sort Shell sort Fancer algorthms: O(n

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

Skew Angle Estimation and Correction of Hand Written, Textual and Large areas of Non-Textual Document Images: A Novel Approach

Skew Angle Estimation and Correction of Hand Written, Textual and Large areas of Non-Textual Document Images: A Novel Approach Angle Estmaton and Correcton of Hand Wrtten, Textual and Large areas of Non-Textual Document Images: A Novel Approach D.R.Ramesh Babu Pyush M Kumat Mahesh D Dhannawat PES Insttute of Technology Research

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

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices Steps for Computng the Dssmlarty, Entropy, Herfndahl-Hrschman and Accessblty (Gravty wth Competton) Indces I. Dssmlarty Index Measurement: The followng formula can be used to measure the evenness between

More information

A Posteriori Multi-Probe Locality Sensitive Hashing

A Posteriori Multi-Probe Locality Sensitive Hashing A Posteror Mult-Probe Localty Senstve Hashng Alexs Joly INRIA Rocquencourt Le Chesnay, 78153, France alexs.joly@nra.fr Olver Busson INA, France Bry-sur-Marne, 94360 obusson@na.fr ABSTRACT Effcent hgh-dmensonal

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

Image Representation & Visualization Basic Imaging Algorithms Shape Representation and Analysis. outline

Image Representation & Visualization Basic Imaging Algorithms Shape Representation and Analysis. outline mage Vsualzaton mage Vsualzaton mage Representaton & Vsualzaton Basc magng Algorthms Shape Representaton and Analyss outlne mage Representaton & Vsualzaton Basc magng Algorthms Shape Representaton and

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

Reducing Frame Rate for Object Tracking

Reducing Frame Rate for Object Tracking Reducng Frame Rate for Object Trackng Pavel Korshunov 1 and We Tsang Oo 2 1 Natonal Unversty of Sngapore, Sngapore 11977, pavelkor@comp.nus.edu.sg 2 Natonal Unversty of Sngapore, Sngapore 11977, oowt@comp.nus.edu.sg

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. 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

Mining User Similarity Using Spatial-temporal Intersection

Mining User Similarity Using Spatial-temporal Intersection www.ijcsi.org 215 Mnng User Smlarty Usng Spatal-temporal Intersecton Ymn Wang 1, Rumn Hu 1, Wenhua Huang 1 and Jun Chen 1 1 Natonal Engneerng Research Center for Multmeda Software, School of Computer,

More information

Face Recognition using 3D Directional Corner Points

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

More information

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

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

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

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

The Codesign Challenge

The Codesign Challenge ECE 4530 Codesgn Challenge Fall 2007 Hardware/Software Codesgn The Codesgn Challenge Objectves In the codesgn challenge, your task s to accelerate a gven software reference mplementaton as fast as possble.

More information

Boundary-Based Time Series Sorting

Boundary-Based Time Series Sorting JOURNAL OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA, VOL. 6, NO. 3, SEPTEMBER 2008 323 Boundary-Based Tme Seres Sortng Jun-Ku L, Yuan-Zhen Wang, and Ha-Bo L Abstract In many applcatons, t s desrable

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

Complex Numbers. Now we also saw that if a and b were both positive then ab = a b. For a second let s forget that restriction and do the following.

Complex Numbers. Now we also saw that if a and b were both positive then ab = a b. For a second let s forget that restriction and do the following. Complex Numbers The last topc n ths secton s not really related to most of what we ve done n ths chapter, although t s somewhat related to the radcals secton as we wll see. We also won t need the materal

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

Random Kernel Perceptron on ATTiny2313 Microcontroller

Random Kernel Perceptron on ATTiny2313 Microcontroller Random Kernel Perceptron on ATTny233 Mcrocontroller Nemanja Djurc Department of Computer and Informaton Scences, Temple Unversty Phladelpha, PA 922, USA nemanja.djurc@temple.edu Slobodan Vucetc Department

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

Vectorization of Image Outlines Using Rational Spline and Genetic Algorithm

Vectorization of Image Outlines Using Rational Spline and Genetic Algorithm 01 Internatonal Conference on Image, Vson and Computng (ICIVC 01) IPCSIT vol. 50 (01) (01) IACSIT Press, Sngapore DOI: 10.776/IPCSIT.01.V50.4 Vectorzaton of Image Outlnes Usng Ratonal Splne and Genetc

More information

Speeding Up the Xbox Recommender System Using a Euclidean Transformation for Inner-Product Spaces

Speeding Up the Xbox Recommender System Using a Euclidean Transformation for Inner-Product Spaces Speedng Up the Xbox Recommender System Usng a Eucldean Transformaton for Inner-Product Spaces Ran Glad-Bachrach Mcrosoft Research Yoram Bachrach Mcrosoft Research Nr Nce Mcrosoft R&D Lran Katzr Computer

More information

Video Content Representation using Optimal Extraction of Frames and Scenes

Video Content Representation using Optimal Extraction of Frames and Scenes Vdeo Content Representaton usng Optmal Etracton of rames and Scenes Nkolaos D. Doulam Anastasos D. Doulam Yanns S. Avrths and Stefanos D. ollas Natonal Techncal Unversty of Athens Department of Electrcal

More information

Discriminative Dictionary Learning with Pairwise Constraints

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

More information

A Workflow for Spatial Uncertainty Quantification using Distances and Kernels

A Workflow for Spatial Uncertainty Quantification using Distances and Kernels A Workflow for Spatal Uncertanty Quantfcaton usng Dstances and Kernels Célne Schedt and Jef Caers Stanford Center for Reservor Forecastng Stanford Unversty Abstract Assessng uncertanty n reservor performance

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

Backpropagation: In Search of Performance Parameters

Backpropagation: In Search of Performance Parameters Bacpropagaton: In Search of Performance Parameters ANIL KUMAR ENUMULAPALLY, LINGGUO BU, and KHOSROW KAIKHAH, Ph.D. Computer Scence Department Texas State Unversty-San Marcos San Marcos, TX-78666 USA ae049@txstate.edu,

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

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

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

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

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