Efficient Mean-shift Clustering Using Gaussian KD-Tree

Size: px
Start display at page:

Download "Efficient Mean-shift Clustering Using Gaussian KD-Tree"

Transcription

1 Pacfc Graphcs 2010 P. Allez, K. Bala, and K. Zhou (Guest Edtors) Volume 29 (2010), Number 7 Effcent Mean-shft Clusterng Usng Gaussan KD-Tree Chunxa Xao Meng Lu The School of Computer, Wuhan Unversty, Wuhan, , Chna Abstract Mean shft s a popular approach for data clusterng, however, the hgh computatonal complexty of the mean shft procedure lmts ts practcal applcatons n hgh dmensonal and large data set clusterng. In ths paper, we propose an effcent method that allows mean shft clusterng performed on large data set contanng tens of mllons of ponts at nteractve rate. The key n our method s a new scheme for approxmatng mean shft procedure usng a greatly reduced feature space. Ths reduced feature space s adaptve clusterng of the orgnal data set, and s generated by applyng adaptve KD-tree n a hgh-dmensonal affnty space. The proposed method sgnfcantly reduces the computatonal cost whle obtanng almost the same clusterng results as the standard mean shft procedure. We present several knds of data clusterng applcatons to llustrate the effcency of the proposed method, ncludng mage and vdeo segmentaton, statc geometry model and tme-varyng sequences segmentaton. Categores and Subject Descrptors (accordng to ACM CCS): I.4 [Computng methodologes]: Image Processng and Computer Vson Applcatons 1. Introducton Mean shft s a well establshed method for data set clusterng, t has been wdely used n mage and vdeo segmentaton [CM02] [WTXC], object trackng [CRM03], mage denosng [BC04], mage and vdeo stylzaton [DS02] [WXSC04], and vdeo edtng [WBC 05], t also has been extended to geometry segmentaton [YLL 05] and 3D reconstructon [WQ04]. Mean shft works by defnng a Gaussan kernel densty estmate for underlyng data, and clusters together the ponts that converge to the same mode under a fxed-pont teratve scheme. Although mean-shft works well for data clusterng and obtan pleasng clusterng results, however, the hgh computatonal complexty s the one of man dffcultes to apply mean shft to cluster large data set, especally for those stuatons where the nteractve and even real tme clusterng processng are preferred. The complexty for the standard mean shft procedure s O(τdn 2 ), where n s the number of the data ponts, the τ s the number of the teratons for mean shft clusterng procedure, and d s the dmenson of the pont. The most expensve computng s to fnd the closest neghborhood for each pont n the data space, whch s a multdmensonal range searchng method. Even usng one of the most popular nearest neghbor search method, the ANN method [AMN 98], gven a query pont q and ε > 0, a (1 + ε) approxmate nearest neghbor of q must be computed n O(c d,ε logn) tme, where c d,ε s a factor dependng on dmenson d and ε. Therefore, when processng large data sets, the hgh tme complexty leads to serous dffculty. Although many acceleraton technques have been proposed [EDD03, GSM03, YDGD03, WBC 05, CP06, PD07, FK09], further mprovements are stll desrable for both performance and clusterng qualty. In ths paper, nspred by the fast hgh-dmensonal flterng method usng Gaussan KD-trees [AGDL09], we propose an effcent paradgm for mean-shft procedure computng. Our method s based on followng key observaton, snce the mean shft procedure clusters those ponts that are feature smlar, whle there are many clusters of ponts whch are hgh smlar n feature, t s wasteful to perform mean shft procedure for each pont to converge to the mode. Thus we frst cluster the orgnal pont set nto clusters based on feature smlarty usng KD-tree [AGDL09], and obtan the Gaussan weghted samples of the orgnal data set, whch s the reduced feature space for approxmatng orgnal pont set. Then nstead of computng mean shft drectly on orgc 2010 The Author(s) Publshed by Blackwell Publshng, 9600 Garsngton Road, Oxford OX4 2DQ, UK and 350 Man Street, Malden, MA 02148, USA.

2 nal ndvdual ponts, we compute on the samples (whch s of a much smaller number) to obtan the modes of the sample space. Fnally we fnd the closest mode for each pont based on Gaussan weghted feature smlarty, and construct the fnal clusterng results. As mean shft s performed on a greatly reduced space (typcally thousands of tmes smaller than orgnal data set), and all stages of our algorthm are data-parallel across queres and can be mplemented the algorthm n CUDA [Buc07], we can cluster the large data set n real tme or at nteractve rate (for a vdeo wth pxels n Fgure 5). Furthermore, as the sample space s an approxmate feature space of the orgnal data set generated usng the proposed Gaussan weghted smlarty samplng, our method receves accurate results comparable wth the standard mean shft that performed on the orgnal data set. In addton, our method uses only an extremely small fracton of resources, for both tme and memory consumng. Ths paper s organzed as follows. In secton 2, we gve the related work, secton 3 s the man part of our paper, we descrbe the proposed fast mean shft clusterng method. In secton 4, we gve the applcatons of our method and comparsons wth the related mean shft acceleraton methods, and we conclude n secton Related work Mean shft was frst presented by Fukunaga and Hostetler [FH75], and t was further nvestgated by Cheng et al. [Che95] and Comancu et al. [CM02]. Mean shft s now a popular approach for data set clusterng, and has been wdely used n mage and vdeo segmentaton [CM02] [WTXC], object trackng [CRM03], mage denosng [BC04] and mage and vdeo stylzaton [DS02] [WXSC04], t also has been extended to geometry segmentaton [YLL 05] and 3D reconstructon [WQ04], and many mage and vdeo edtng methods are based on the mean shft clusterng preprocessng [WBC 05]. One of the man dffcultes n applyng Mean Shft based clusterng to large data sets s ts computatonal complexty. For each Gaussan weghted average teraton, the complexty of brute force computaton s quadratc n the number of data ponts. There are several exstng technques whch have been developed to ncrease the speed of Mean Shft. Comancu and Meer [CM02] used axs-algned box wndows, however, ths produces many lmt ponts and adjacent ponts are merged as a post-process. Dementhon [DeM02] used multscale structure to accelerate vdeo segmentaton. Yang at al. [YDDD03] appled the Fast Gauss Transform to speed up the sums of Gaussans n hgher dmensons that were used n the Mean Shft teraton. Ths method s effectve for Gaussan weghted average wth large flterng radus, however, performng weghted average n a relatve small radus does not beneft much from ths method. Georgescu et al. [GSM03] accelerated mean shft by performng fast nearest neghbor search wth spatally coherent hash tables. Carrera-Perpnán [CP06] studed four acceleraton strateges and found that spatal dscretzaton method (usng unform down samplng schemes) performed best. Pars and Durand [PD07] appled the sparse representaton of the densty functon to accelerate mean shft, smlar to the blateral flterng [PD06], they frst bnned the feature ponts n a coarse regular grd, and then blurred the bn values usng a separable Gaussan. The computatonal complexty and memory scale exponentally wth the dmenson d. Wang et al. [WLGR07] used a dual tree to speed up Mean Shft by computng two separate trees, one for the query ponts, and one for the reference ponts. Compared to the methods of [YDDD03] [PD07], ths method mantans a relatve error bound of the Mean Shft teraton at each stage, leadng to a more accurate algorthm, however, the performance of ths method s much lower than [YDDD03] [PD07]. More recently, Freedman and Kslev [FK09] proposed a samplng technque for Kernel Densty Estmate (KDE), they constructed a compactly represented KDE wth much smaller descrpton complexty, ths method greatly accelerates the mean shft procedure, however, the accuracy of the mean shft clusterng depends on the number of the random samples. Many methods have appled Gaussan KD-Trees for acceleratng mage and vdeo processng. Adams et al. [AGDL09] appled Gaussan KD-Trees for acceleratng hgh-dmensonal flterng ncludes the blateral mage flter [TM98], blateral geometry flterng [JDD03, FDCO03] and mage denosng wth nonlocal means [BCM05]. We borrow some deas from [AGDL09] for adaptve clusterng n ths paper. Xu et al. [XLJ 09] used K-D Tree to buld adaptve clusterng for acceleratng affnty-based mage and vdeo edt propagaton. As an alteratve, Xao et al. [XNT10] used quadtree based herarchcal data structure to accelerate edt propagaton. KD-Trees have been wdely used n acceleratng graphcs renderng [HSA91], the real-rme KDtree constructon on graphcs hardware also have been proposed [HSHH07] [ZHWG08]. Our method apples Gaussan KD-Trees [AGDL09] to buld a herarchy and clusterng for the large data set to accelerate the mean shft computng. Compared wth Pars and Durand [PD07], whose complexty s exponental n the dmenson d of the pont, our tree-based mean shft provdes wth excellent performance, as ts runtme and memory consumng both scale lnearly wth dmenson of the ponts. Wth the samples generated usng smlarty-based KD-tree clusterng, our method obtans more accurate results than [FK09] when usng smlar number of samples. 3. Fast mean shft clusterng We frst gve bref revew of mean shft, then we descrbe the proposed fast mean shft clusterng method, ncludng

3 data set clusterng preprocess usng KD-tree, sample feature space computng, mean shft modes computng n reduced feature space, modes nterpolaton. We also present the complexty analyss and GPU mplementaton of the proposed algorthm Revew of mean shft Gven pont data set {x } n =1, where x R d s d dmensonal feature vector, each s assocated wth a bandwdth value h > 0. An adaptve nonparametrc estmator of ths data set s defned as ( f K (x) = 1 n 1 n =1 h d k x x 2) (1) h ( where K (x) = c k,d k x 2) > 0 s kernel functon satsfyng K (x) 0 and Rd K (x)dx = 1, By takng the gradent of (1) the followng expresson can be obtaned. f K (x) = 2c n n =1 1 h d g ( x x h 2 ) m(x) (2) where g(x) = k (x), and m(x) s the so-called mean shft vector n 1 x x =1 x g( 2 ) h m(x) = n =1 h d+2 1 h d+2 g( x x h 2 ) x (3) The expresson (3) shows that at locaton x the weghted average of the data ponts selected wth kernel K s proportonal to the normalzed densty gradent estmate obtaned wth kernel K. The mean shft vector thus ponts toward the drecton of maxmum ncrease n the densty. The followng gradent-ascent process wth an adaptve step sze untl convergence consttutes the core of the mean shft clusterng procedure n 1 x+1 =1 h d+2 = n 1 =1 g h d+2 ( x x g x ( x x 2 ) h ), j = 1,2,... (4) 2 h The startng pont of the procedure x can be chosen as ponts x, and the convergence ponts of the teratve procedure are the modes of the densty. The all ponts that converge to the same mode are collected and consdered as a cluster. More detals are descrbed n [CM02] Fast mean shft computng The weghted average of expresson (4) s the most tme consumng computng of mean-shft when the number n of the data set s large (for example, 10 9 pxel n vdeo streamng). Gven an arbtrary set of pont {x } n =1 wth feature vector of d dmenson, a nave computaton of mean shft vector expresson (4) would take O(dn 2 ) tme, as every pont nteracts wth every other pont. A smple way to accelerate the mean shft procedure s usng the weghted average of the closest ponts of x, and the bandwdth value h can be set dependng on the neghborhood sze. However, usng ths scheme, we have to perform the nearest neghborhood search, whch s also a tme consumng operaton for large data set, especally for data set wth hgh dmenson vector. To accelerate the weghted average operaton of expresson (4), nstead of computng expresson (4) for ndvdual ponts n the data set, we approxmate orgnal data set by pece-wse lnear segments n the feature space based on smlarty measure, each represented by a cluster of nearby pxels, and the sze of each cluster s adapted to the sze of the smlar feature space. The generated clusters can be consdered as the samples of the data set, whch s of a much smaller number than the number of pont. Then nstead of solvng the mean shft procedure (4) drectly on ndvdual ponts as done n prevous methods, we solve t on the samples based on Gaussan weghted average on a neghborhood, fnally we nterpolate the clusterng results to the orgnal data set. We cluster the pont data based on smlarty measure between the ponts, whch s defned n the feature space of nput data set. We defne the smlarty between the ponts usng both spatal localty p and value v of pont, whch consttutes the feature space of the nput data set. For example, n mage case, pont x s a pxel wth ts poston p = (x,y) and ts color value v = (r,g,b) (Lab color space). Thus, each pont x s a fve-dmensonal feature vector whose axes are x = (p,v) = (x,y,r,g,b). As stated n [AP08] [XLJ 09], the smlarty measure ((or affnty) between two ponts can be defned as z j = exp x x j 2 ), the poston p and value v also can be weghted by parameters. For vdeo, the feature vector can be expanded to nclude the frame ndex t and moton estmaton ς of pont x, and feature vector s expressed as seven-dmensonal vector x = (p,v,t,ς). For mage, we compute the mean shft clusterng procedure (4) n feature space where each pont s assocated wth both spatal localty p and value v KD-Tree adaptve clusterng We apply KD-tree to adaptvely cluster the data set n the feature space, and subdvde fnely n the regons where the pont feature vectors are dfferent, subdvde coarsely n the regons where the feature vectors are smlar. In mage KDtree clusterng, for example, we subdvde the homogeneous regons coarsely, whle subdvde the edges regons fnely. Then by representng each cluster wth a sample, we can receve an accurate approxmate feature space of the orgnal data set wth much less samples. KD-tree clusters the data set based on feature space n a top down way. Startng from a root cell, the top rectangular

4 cell represents all ponts n the data set, we recursvely splt a cell to two chld cells adaptvely along a dmenson that s alternated at successve tree levels. Smlar to [AGDL09], each nner cell of the tree T represents a d-dmensonal rectangular cell whch stores sx varables: the dmenson d along whch the tree cuts, the cut value T cut on that dmenson, the bounds of the cell n that dmenson T mn and T max, and ponters to ts chldren T le ft and T rght. Leaf nodes contan only a d-dmensonal pont whch s consdered as a sample. Ths sample represents the ponts that contaned n the leaf cell, the kd-tree stores m samples { } m y j of orgnal data set { n } d-dmensons, one sample pont per leaf. The samples m y j construct the reduced feature space of the orgnal data set. As llustrated n Fgure 1, we adaptvely cluster the mage nto clusters. The mage s adaptvely clustered, where at the edge regons, more samples are placed, whle at the homogeneous regons, coarse samples are placed. The sample factor can be changed by the stoppng crtera. We present two thresholds for stoppng crtera, one s the sze of the cell, other one s the varance σ of the smlarty measure z j between the ponts n the cell. By usng these two thresholds we can generate dfferent samplng factor for mage as well as respectng for the feature dstrbuton of the data set. Fgure 1 shows some cluster results wth dfferent samplng factor. Fgure 1: Image samplng usng adaptve KD-tree. (a) Orgnal mage ( ), (b) mage clusterng wth 1,050 samples, (c) mage clusterng wth 11,305 samples Sample feature space computng We obtan the samples { } m y j n d-dmensons, whch are adaptve samples of the feature space of orgnal data set. The samples { } m y j can be consdered a approxmate feature space of orgnal space. To make a more accurate ap- proxmate feature space, we scatter the pont x to the samples { } m y j based on the affnty smlarty z j between the pont x and sample y j, and obtan an affnty smlarty based sample space. Smlar to the splattng stages n [AGDL09], we scatter { each } pont x to ts nearest neghborhood N (x ) n samples m y j (We apply the KD-tree to search the hgh dmenson nearest neghborhood N (x ) for pont x ), and compute an affnty based sample y j for each sample y. The affnty based sample y j can be consdered as the weghted smlarty average for those feature vector {x } that s most smlar to the sample y. We compute and sum the affnty smlarty z j between x and each y j N (x ) to obtan the affnty based sample y j of the sample y : y j = y j + z j x, then y j s normalzed by the sum of the smlarty z j. The generated y j s a feature vector of d-dmensons. The affnty based sample {y j } m s a more accurate samples of the orgnal pont set, and wll be used n the mean shft clusterng procedure and modes nterpolaton. Then for each sample, we store two vectors, one s feature vector y, the other s the affnty based sample y j Mean-shft modes computng After obtanng the affnty based samples {y j } m for orgnal data set feature space, nstead of computng the mean shft clusterng procedure n the orgnal space, we compute the mean shft procedure on the reduced feature space {y j } m. Note for each teraton, we fnd for each sample y j the nearest neghborhood N ( y ) j n the sample space, and perform followng gradent-ascent process wth an adaptve step sze untl convergence: y N(y j ) y u g( j y h 2) u j+1 = ( y N(y j ) g u j y h 2), j = 1,2,... (5) Iteratng an nfnte number of tmes the expresson (5) s guaranteed to brng u j to the mode n reduced feature space. Practcally, n the mplementaton, we fnd that the meanshft clusterng procedure (5) tends to converge n a very small number of steps, typcally around 6. As the number of the samples { y } m j s much smaller than the number of the ponts {x } n =1, that s m n, the computatonal complexty of mean shft vector computng has been reduced sgnfcantly. Furthermore, the mean shft s performed n the affnty smlarty based reduced feature space, whch leads to more accurate modes. We apply the KD-tree to perform the hgh dmenson nearest neghbor research for each teraton computng. After performng mean shft teratons for each sample y j, we receve the modes {z k } s k=1 of the reduced feature space { y } m j, whch are the approxmate modes of orgnal data set. All samples convergng to the same mode are clustered together Modes nterpolaton To nterpolate the modes computed n the reduced feature space to the orgnal data set, one naïve approach s to fnd a nearest mode z l {z k } s k=1 for each pont x. Ths can be consdered as a hard clusterng. As an alternatve method, we gve a soft clusterng method whch generates more smooth clusterng results. Ths method works by applyng weghted based nterpolaton.

5 The mode nterpolaton works as follows, for each pont x, we fnd the nearest samples N (x ) n { y } m j. Each sample y j N (x ) converges to a mode u j, that s, y j u j. Based on the affnty smlarty z j between x and y j, by normalzng the weghts z j : j z j = 1, the fnal mode are computed as: x z j u j. When we compute the weghted mode nterpolaton over all the samples { y } m j, smlar to [FK09], we wll receve the cartoon-lke clusterng results. Note that n the nterpolaton stage, the sze of the N (x ) s not always the same as that performed n the sample space computng stage. In our experment, we set neghborhood sze between 10 and 20, and receve satsfed results. As the samples are sgnfcantly smaller than the number of the orgnal pont set, usng the GPU accelerated KD-tree nearest neghborhood search, the search performng s fast. In addton, snce we determne the fnal mode for x based on the weghted smlarty, the results are more accurate. Fgure 3 shows the results usng the two dfferent mode selecton methods, one s the nearest-sample based mode selecton, the other s weghted modes nterpolaton. As llustrated n Fgure 3, usng the weghted modes nterpolaton, we receve smoother and more accurate results. up-sample clusterng results to the orgnal data sets whch takes O(n(logm + d)) tme. Recall that m n, ths results n a total complexty O(dnlogn). Compared wth standard mean shft procedure wth complexty O(τdn 2 ), the proposed method sgnfcantly accelerated. Applyng the method presented by [ZHWG08], a KD-tree s effcently bult on the GPU for the nput ponts wth hghdmensonal feature vector. Three stages of our proposed algorthm, ncludng the ponts scatterng, mean shft clusterng procedure n the reduced space, and modes nterpolaton, all ncorporate the hgh dmensonal nearest neghbor search. As the hgh dmensonal nearest neghbor search can be mplemented n parallel usng GPU [AGDL09], thus, our method s even fast to process large data set wth hgh dmensonal feature vector. We mplement the proposed algorthms n CUDA, and run t on an NVIDIA GeForce GTX 285 (1GB) graphcs card. We observe a typcal speedup of 20x over our sngle-threaded CPU mplementaton runnng on an Pentum(R) Dual-Core CPU E5200@2.50GHz wth 2GB RAM, whch allows our method to be appled n an nteractve mean shft clusterng framework for moderateszed data set. 4. Applcatons and comparsons (a) (b) (c) Fgure 2: Image segmentaton results comparson. (a) The orgnal mage, (b) the result usng nearest sample based mode selecton, (c) the result usng weghted modes nterpolaton Complexty analyss and GPU mplementaton Our algorthm accelerates the mean shft procedures by computng a lower resoluton feature space and then nterpolatng the clusterng result to the orgnal data set. By usng a KD-tree structure, we construct a reduced feature space for nput n d-dmensonal data ponts wth m feature vector: {x } n =1 { y } m j, and m n. Assumng the depth of Gaussan tree s O(logm), the complexty of tree constructon s O(ndlog m). Performng nearest neghborhood for each of the n nput ponts to scatter values nto the tree takes O(n(logm + d)) tme. If performng τ teratons for mean shft clusterng procedure, computng the mean shft teraton n the reduced space wth m feature vector takes O(τm(logm + d)) tme. In the last stage, we We apply the proposed fast mean shft clusterng to followng applcatons, mage segmentaton, vdeo segmentaton, geometry model segmentaton, and anmated object segmentaton. We also present the comparson results on both performance and segmentaton qualty wth the most related methods. Our approach s mplemented usng C++ on a machne equpped wth Pentum(R) Dual-Core CPU E5200@2.50GHz wth 2GB RAM. The GPU acceleraton s based on CUDA [ http: ] and run on a NVIDIA GeForce GTX 285 (1GB) graphcs card Image segmentaton We apply the proposed fast mean shft method for mage segmentaton. All pxels that converge to same mode are collected together and are consdered to be the same segment. In mage case, we defne the feature vector of pxel x = (σ p p,σ cc ) comprsng ts poston p = (x,y) and ts color value c = (r,g,b) (Lab space) whch can be weghted by parameters σ p and σ c. Thus, each pxel x s a fvedmensonal vector. Fgure 3 presents the segmentaton results generated applyng our fast mean shft method based on dfferent samplng factor. As llustrated n Fgure 3, there are pxels n the orgnal mage. Even wth very hgh samplng factor such as n/m = 4, 096, the segmentaton results s stll pleasng. Wth much less samples, the mage can be clustered n hgh speed even wthout usng GPU acceleraton. It

6 takes only total seconds on CPU to perform mean shft clusterng wth samplng factor n/m = 1,024. In Fgure 4, we present mage segmentaton results for the mages wth dfferent szes, and gve the comparson results wth standard mean shft method [CM02], the accelerated method of Pars and Durand [PD07], and compactly represented KDE of Freedman and Kslev [FK09]. We comprse wth these methods on both performance and segmentaton qualty. Compared wth [CM02], some weak features may be lost usng our method snce they may be ncorporated nto the salent features durng data Gaussan KD-tree clusterng, however, the salent features may be better kept, as llustrated n Fgure 4. As shown n Fgure 4, gven the same samplng factor, our method generates hgher qualty compared wth [PD07] and [FK09], especally at the regons wth weak edges. The complexty of [PD07] depends on the dmenson d of the pont, when processng hgh-dmensonal data, ths method does not show much advantage. However, our method s fast even wth low samplng factor and hgh dmensonal data sets. as shown n Table 1. It takes our method 5.91 second to cluster pxels on CPU, whle t take seconds usng [PD07]. Usng our method ncorporatng GPU mplementaton, our method shows greater advantage when processng large data sets wth hgh dmensonal feature vector, t takes less than 0.2 second to cluster pxels on GPU Vdeo segmentaton Mean shft clusterng can also be used n vdeo segmentaton [DeM02]. As vdeo streamng usually contans mllons of pxels, practcal vdeo segmentaton usng mean shft clusterng depends heavly on the performance of the mean shft procedure. In addton, compared wth mage data, the dmensons of feature space are hgher, whch further ncrease the computatonal complexty of mean shft procedure. Thus, t s mpractcable to segment long range vdeo streamng by performng standard mean shft clusterng wthout usng acceleraton technques. However, usng our fast mean shft clusterng method, we can perform vdeo segmentaton at nteractve rate. We defne a feature space {x } n =1 of seven dmensons for the vdeo. The feature vector at each pxel x = (σ p p,σ c c,σ t t,σ ς ς ) comprses ts poston p (x and y coordnate), color c (Lab color vector), tme t and moton ς, these four knds of features can be weghted by parameters σ p, σ v, σ t and σ ς, and the values of these parameters are defned as constants for all pxels. As llustrated n Fgure 5, there are pxels n the vdeo, and we frst cluster the vdeo wth samplng factor usng KD tree. It takes our method seconds to perform the mean shft clusterng on CPU, and 1.2 seconds on GPU. It takes seconds on CPU usng [PD07]. Fgure 5: Vdeo segmentaton. Left column, nput vdeo ( ), from top to down, 1th frame, 28th frame, 50th frame. Rght column, segmentaton results Mesh model segmentaton Smlar to mage and vdeo segmentaton n mage processng and analyss, surface segmentaton s one of the most mportant operatons n geometry processng and modelng. Yamauch et al. [YLL 05] adopted the mean shft clusterng to mesh partton, and produced feature senstve mesh segmentaton. In Fgure 6, we gve the mesh segmentaton usng the proposed fast mean shft clutterng. We defne a feature space {x } n =1 of sx dmensons for the mesh model. The feature vector at vertex x = (σ p p,σ v v ) comprses ts poston p (x, y and z coordnate) and normal v (three dmenson vector) whch can be weghted by parameters σ p and σ v. The values of σ p and σ v are defned as global constants for all vertces. In Fgure 6, usng the varant of mean shft procedure to the mesh model, we receve a patch-type segmentaton results, and the segmentaton results are senstve to the salent surface features. Furthermore, we adopt the herarchcal mage segmentaton method [PD07] to mesh model and generate the herarchcal segmentaton results. Note that our fast mean shft method s guaranteed to produce segmentaton results whch catch the meanngful components, no addtonal computaton s needed to compute the herarchcal results. For a mesh model wth 70,994 vertces, t takes 0.31 second for our method to compute the results. We also gve the comparson results wth [YLL 05]. As shown n Fgure 6, our approach generates more convncng results Anmated geometry object segmentaton The proposed fast mean shft also can be used to accelerate the anmated object (geometry surface sequences) segmenc 2010 The Author(s)

7 (a) (b) (c) (d) (e) Fgure 3: Image segmentaton usng dfferent samplng factor. (a) Orgnal mage, (b) n/m =256, (c) n/m =1,024, (d) n/m =4,096, (e) n/m =16,384. Data set Data set sze d n/m Kd-tree constructon Modes computng Modes nterpolaton tme brd Obama castle Vdeo Mesh Horse Table 1: Performance of our method for dfferent knds of data sets. Note that we perform clusterng for anmated object (Horse) on GPU. All other data sets are performed on CPU. (a) (b) (c) (d) (e) (f) Fgure 6: Herarchcal decomposton of statc mesh model. (a)-(e) s the results of our proposed herarchcal decomposton method, (f) s the result of [YLL 05]. taton. Inspred by [LXL ], we frst compute approxmately nvarant sgnature vectors ξ for each vertex of the anmated object [LG05], whch s a local and hgh dmensonal approxmately nvarant under shape rgd/scalng transformatons. Then both the geometrc attrbutes (vertex poston p and ts normal v ) and ts local sgnature vector ξ of each vertex x on the anmated object can be weghted by parameters σ p, σ v and σ ξ, whch construct the hgh dmensonal feature space of the anmated object x = (σ p p,σ vv,σ ξ ξ ). Then the vertces of anmated object can be clustered effcently usng the proposed GPU- accelerated mean shft clusterng algorthm. In Fgure 7, we gve the anmated object segmentaton results. There are total vertces n the anmated object wth 50 frames. We use d = 24 dmensons for the sgnature vector ξ (ξ R 24 ) n our mplementaton. It takes 1.5 seconds on GPU to complete the mean shft teratons (10 teratons n ths example) wth samplng factor n/m = We also gve the comparson results wth [WB10]. Wuhrer and Brunton [WB10] performed the anmated object segmentaton n dual space of the mesh model. They found near-rgd segments whose segment boundares locate at regons of large deformaton, and assumed that the vertex-tovertex correspondences of the nput meshes were known. As an alternatve, our method reles on the local hgh dmensonal sgnature vector nformaton for clusterng, ncorporatng wth the proposed fast mean shft clusterng technques, whch ensures the decomposed parts more meanngful and temporally-coherent results n hgher speed. 5. Concluson In ths paper, we propose a new algorthm for acceleratng compute mean shft clusterng. Usng KD-tree to adaptvely cluster the orgnal data set nto clusters wth smlar feature smlarty, the clusters construct the samples of the orgnal data set. Then we compute the mean shft procedure on

8 (a) (b) (c) (d) (e) Fgure 4: Image segmentaton comparsons. (a) orgnal mage, (b) mage segmentaton usng standard mean shft (EDISON), (c) mage segmentaton usng [PD07], (d) mage segmentaton usng [FK09], (e) mage segmentaton usng our proposed method. the greatly reduced sampled feature space and generated the modes, and fnally by usng the Gaussan mportance weght, we upsample the computed modes to the orgnal data set to get fnal clusterng results. Our algorthm sgnfcantly speeds up the performance whle not sacrfcng the accuracy. Our method s especally useful for hgh resoluton mages, long tme vdeo sequences and geometry models segmentaton wth large pont set. Acknowledgment We would lke to thank the anonymous revewers for ther valuable comments and nsghtful suggestons. Ths work was partly supported by NSFC (No ), Natonal Hgh Technology Research and Development Program of Chna (863 Program) (No. 2008AA121603), Natural Scence Foundaton of Hube Provnce (2008CDB350), State Key Lab of CAD&CG (No. A0808), the Fundamental Research Funds for the Central Unverstes ( ). References [AGDL09] ADAMS A., GELFAND N., DOLSON J., LEVOY M.: Gaussan kd-trees for fast hgh-dmensonal flterng. ACM Transactons on Graphcs (TOG) 28, 3 (2009), 21. [AMN 98] ARYA S., MOUNT D., NETANYAHU N., SILVER- MAN R., WU A.: An optmal algorthm for approxmate nearc 2010 The Author(s)

9 Fgure 7: Anmated object decomposton result comparson. Top row: our results, bottom row: Wuhrer and Brunton [WB10]. est neghbor searchng fxed dmensons. Journal of the ACM (JACM) 45, 6 (1998), [AP08] AN X., PELLACINI F.: Appprop: all-pars appearancespace edt propagaton. ACM Trans. Graph 27, 3 (2008), 40. [BC04] BARASH D., COMANICIU D.: A common framework for nonlnear dffuson, adaptve smoothng, blateral flterng and mean shft. Image and Vson Computng 22, 1 (2004), [BCM05] BUADES A., COLL B., MOREL J.: A non-local algorthm for mage denosng. In CVPR 2005 (2005), pp [Buc07] BUCK I.: Gpu computng: Programmng a massvely parallel processor. In Proceedngs of the Internatonal Symposum on Code Generaton and Optmzaton (2007), IEEE Computer Socety, p. 17. [Che95] CHENG Y.: Mean shft, mode seekng, and clusterng. IEEE Transactons on Pattern Analyss and Machne Intellgence 17, 8 (1995), [CM02] COMANICIU D., MEER P.: Mean shft: A robust approach toward feature space analyss. IEEE Transactons on pattern analyss and machne ntellgence 24, 5 (2002), [CP06] CARREIRA-PERPINÁN M.: Acceleraton strateges for Gaussan mean-shft mage segmentaton. In CVPR (2006), vol. 1. [CRM03] COMANICIU D., RAMESH V., MEER P.: Kernel-based object trackng. IEEE Transactons on Pattern Analyss and Machne Intellgence 25, 5 (2003), [DeM02] DEMENTHON D.: Spato-temporal segmentaton of vdeo by herarchcal mean shft analyss. Language 2 (2002). [DS02] DECARLO D., SANTELLA A.: Stylzaton and abstracton of photographs. In SIGGRAPH (2002), ACM, pp [EDD03] ELGAMMAL A., DURAISWAMI R., DAVIS L.: Effcent kernel densty estmaton usng the fast gauss transform wth applcatons to color modelng and trackng. IEEE Transactons on Pattern Analyss and Machne Intellgence 25, 11 (2003), [FDCO03] FLEISHMAN S., DRORI I., COHEN-OR D.: Blateral mesh denosng. ACM Transactons on Graphcs (TOG) 22, 3 (2003), [FH75] FUKUNAGA K., HOSTETLER L.: The estmaton of the gradent of a densty functon, wth applcatons n pattern recognton. IEEE Transactons on Informaton Theory 21, 1 (1975), [FK09] FREEDMAN D., KISILEV P.: Fast Mean Shft by compact densty representaton. [GSM03] GEORGESCU B., SHIMSHONI I., MEER P.: Mean shft based clusterng n hgh dmensons: A texture classfcaton example. In ICCV (2003), pp [HSA91] HANRAHAN P., SALZMAN D., AUPPERLE L.: A rapd herarchcal radosty algorthm. ACM SIGGRAPH Computer Graphcs 25, 4 (1991), 206. [HSHH07] HORN D., SUGERMAN J., HOUSTON M., HANRA- HAN P.: Interactve kd tree GPU raytracng. In Proceedngs of the 2007 symposum on Interactve 3D graphcs and games (2007), ACM, p [JDD03] JONES T., DURAND F., DESBRUN M.: Non-teratve, feature-preservng mesh smoothng. ACM Transactons on Graphcs 22, 3 (2003), [LG05] LI X., GUSKOV I.: Mult-scale features for approxmate algnment of pont-based surfaces. In SGP (2005), Eurographcs Assocaton, p [LXL ] LIAO B., XIAO C., LIU M., DONG Z., PENG Q.: Fast Herarchcal Anmated Object Decomposton Usng Approxmately Invarant Sgnature. Submmted to The Vsual Computer. [PD06] PARIS S., DURAND F.: A fast approxmaton of the blateral flter usng a sgnal processng approach. ECCV (2006). [PD07] PARIS S., DURAND F.: A topologcal approach to herarchcal segmentaton usng mean shft. In CVPR (2007), pp [TM98] TOMASI C., MANDUCHI R.: Blateral flterng for gray and color mages. [WB10] WUHRER S., BRUNTON A.: Segmentng anmated objects nto near-rgd components. The Vsual Computer 26, 2 (2010), [WBC 05] WANG J., BHAT P., COLBURN R., AGRAWALA M., COHEN M.: Vdeo cutout. ACM Transactons on Graphcs 24, 3 (2005), [WLGR07] WANG P., LEE D., GRAY A., REHG J.: Fast mean shft wth accurate and stable convergence. In Workshop on Artfcal Intellgence and Statstcs (AISTATS) (2007), Cteseer. [WQ04] WEI Y., QUAN L.: Regon-based progressve stereo matchng. In CVPR (2004), vol. 1, Cteseer, pp [WTXC] WANG J., THIESSON B., XU Y., COHEN M.: Image and vdeo segmentaton by ansotropc kernel mean shft. Computer Vson-ECCV 2004, [WXSC04] WANG J., XU Y., SHUM H., COHEN M.: Vdeo toonng. In ACM SIGGRAPH 2004 Papers (2004), ACM, pp [XLJ 09] XU K., LI Y., JU T., HU S., LIU T.: Effcent affntybased edt propagaton usng KD tree. In ACM SIGGRAPH Asa 2009 papers (2009), ACM, pp [XNT10] XIAO C., NIE Y., TANG F.: Effcent Edt Propagaton Usng Herarchcal Data Structure. IEEE Transactons on Vsualzaton and Computer Graphcs (2010). [YDDD03] YANG C., DURAISWAMI R., DEMENTHON D., DAVIS L.: Mean-shft analyss usng quas-newton methods. In ICIP (2003), vol. 3, Cteseer, pp [YDGD03] YANG C., DURAISWAMI R., GUMEROV N., DAVIS L.: Improved fast gauss transform and effcent kernel densty estmaton. In ICCV (2003), pp [YLL 05] YAMAUCHI H., LEE S., LEE Y., OHTAKE Y., BELYAEV A., SEIDEL H.: Feature senstve mesh segmentaton wth mean shft. In SMA (2005), vol. 243, IEEE. [ZHWG08] ZHOU K., HOU Q., WANG R., GUO B.: Real-tme kd-tree constructon on graphcs hardware. In ACM SIGGRAPH Asa 2008 papers (2008), ACM, pp

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

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

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

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

Simplification of 3D Meshes

Simplification of 3D Meshes Smplfcaton of 3D Meshes Addy Ngan /4/00 Outlne Motvaton Taxonomy of smplfcaton methods Hoppe et al, Mesh optmzaton Hoppe, Progressve meshes Smplfcaton of 3D Meshes 1 Motvaton Hgh detaled meshes becomng

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

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

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

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

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

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

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

THE PULL-PUSH ALGORITHM REVISITED

THE PULL-PUSH ALGORITHM REVISITED THE PULL-PUSH ALGORITHM REVISITED Improvements, Computaton of Pont Denstes, and GPU Implementaton Martn Kraus Computer Graphcs & Vsualzaton Group, Technsche Unverstät München, Boltzmannstraße 3, 85748

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

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

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

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

An Image Fusion Approach Based on Segmentation Region

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

More information

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

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

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

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

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

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

Feature-Preserving Mesh Denoising via Bilateral Normal Filtering

Feature-Preserving Mesh Denoising via Bilateral Normal Filtering Feature-Preservng Mesh Denosng va Blateral Normal Flterng Ka-Wah Lee, Wen-Png Wang Computer Graphcs Group Department of Computer Scence, The Unversty of Hong Kong kwlee@cs.hku.hk, wenpng@cs.hku.hk Abstract

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

A fast algorithm for color image segmentation

A fast algorithm for color image segmentation Unersty of Wollongong Research Onlne Faculty of Informatcs - Papers (Arche) Faculty of Engneerng and Informaton Scences 006 A fast algorthm for color mage segmentaton L. Dong Unersty of Wollongong, lju@uow.edu.au

More information

SIGGRAPH Interactive Image Cutout. Interactive Graph Cut. Interactive Graph Cut. Interactive Graph Cut. Hard Constraints. Lazy Snapping.

SIGGRAPH Interactive Image Cutout. Interactive Graph Cut. Interactive Graph Cut. Interactive Graph Cut. Hard Constraints. Lazy Snapping. SIGGRAPH 004 Interactve Image Cutout Lazy Snappng Yn L Jan Sun Ch-Keung Tang Heung-Yeung Shum Mcrosoft Research Asa Hong Kong Unversty Separate an object from ts background Compose the object on another

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

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

Wavefront Reconstructor

Wavefront Reconstructor A Dstrbuted Smplex B-Splne Based Wavefront Reconstructor Coen de Vsser and Mchel Verhaegen 14-12-201212 2012 Delft Unversty of Technology Contents Introducton Wavefront reconstructon usng Smplex B-Splnes

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

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

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

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

Adaptive Selection of Rendering Primitives for Point Clouds of Large Scale Environments

Adaptive Selection of Rendering Primitives for Point Clouds of Large Scale Environments Adaptve Selecton of Renderng Prmtves for Pont Clouds of Large Scale Envronments Taash Maeno, Hroa Date and Satosh Kana 3 Graduate School of Informaton Scence and Technology, Hoado Unversty, Japan t_maeno@sdm.ss.st.houda.ac.jp,

More information

Robust Mean Shift Tracking with Corrected Background-Weighted Histogram

Robust Mean Shift Tracking with Corrected Background-Weighted Histogram Robust Mean Shft Trackng wth Corrected Background-Weghted Hstogram Jfeng Nng, Le Zhang, Davd Zhang and Chengke Wu Abstract: The background-weghted hstogram (BWH) algorthm proposed n [] attempts to reduce

More information

Mesh Editing in ROI with Dual Laplacian

Mesh Editing in ROI with Dual Laplacian Mesh Edtng n ROI wth Dual Laplacan Luo Qong, Lu Bo, Ma Zhan-guo, Zhang Hong-bn College of Computer Scence, Beng Unversty of Technology, Chna lqngng@sohu.com, lubo@but.edu.cn,mzgsy@63.com,zhb@publc.bta.net.cn

More information

COMPLEX WAVELET TRANSFORM-BASED COLOR INDEXING FOR CONTENT-BASED IMAGE RETRIEVAL

COMPLEX WAVELET TRANSFORM-BASED COLOR INDEXING FOR CONTENT-BASED IMAGE RETRIEVAL COMPLEX WAVELET TRANSFORM-BASED COLOR INDEXING FOR CONTENT-BASED IMAGE RETRIEVAL Nader Safavan and Shohreh Kasae Department of Computer Engneerng Sharf Unversty of Technology Tehran, Iran skasae@sharf.edu

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

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

High resolution 3D Tau-p transform by matching pursuit Weiping Cao* and Warren S. Ross, Shearwater GeoServices

High resolution 3D Tau-p transform by matching pursuit Weiping Cao* and Warren S. Ross, Shearwater GeoServices Hgh resoluton 3D Tau-p transform by matchng pursut Wepng Cao* and Warren S. Ross, Shearwater GeoServces Summary The 3D Tau-p transform s of vtal sgnfcance for processng sesmc data acqured wth modern wde

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

Feature-Area Optimization: A Novel SAR Image Registration Method

Feature-Area Optimization: A Novel SAR Image Registration Method Feature-Area Optmzaton: A Novel SAR Image Regstraton Method Fuqang Lu, Fukun B, Lang Chen, Hao Sh and We Lu Abstract Ths letter proposes a synthetc aperture radar (SAR) mage regstraton method named Feature-Area

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

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

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

MOTION BLUR ESTIMATION AT CORNERS

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

More information

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Proceedngs of the Wnter Smulaton Conference M E Kuhl, N M Steger, F B Armstrong, and J A Jones, eds A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Mark W Brantley Chun-Hung

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

Active Contours/Snakes

Active Contours/Snakes Actve Contours/Snakes Erkut Erdem Acknowledgement: The sldes are adapted from the sldes prepared by K. Grauman of Unversty of Texas at Austn Fttng: Edges vs. boundares Edges useful sgnal to ndcate occludng

More information

Fitting: Deformable contours April 26 th, 2018

Fitting: Deformable contours April 26 th, 2018 4/6/08 Fttng: Deformable contours Aprl 6 th, 08 Yong Jae Lee UC Davs Recap so far: Groupng and Fttng Goal: move from array of pxel values (or flter outputs) to a collecton of regons, objects, and shapes.

More information

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning

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

More information

Fast Integral MeanShift : Application to Color Segmentation of Document Images

Fast Integral MeanShift : Application to Color Segmentation of Document Images 2013 12th Internatonal Conference on Document Analyss and Recognton Fast Integral MeanShft : Applcaton to Color Segmentaton of Document Images Frank LeBourgeos*, Fadoua Drra**, Djamel Gaceb*, Jean Duong*

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

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

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

SCALABLE AND VISUALIZATION-ORIENTED CLUSTERING FOR EXPLORATORY SPATIAL ANALYSIS

SCALABLE AND VISUALIZATION-ORIENTED CLUSTERING FOR EXPLORATORY SPATIAL ANALYSIS SCALABLE AND VISUALIZATION-ORIENTED CLUSTERING FOR EXPLORATORY SPATIAL ANALYSIS J.H.Guan, F.B.Zhu, F.L.Ban a School of Computer, Spatal Informaton & Dgtal Engneerng Center, Wuhan Unversty, Wuhan, 430079,

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

An Entropy-Based Approach to Integrated Information Needs Assessment

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

More information

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

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

More information

Some Tutorial about the Project. Computer Graphics

Some Tutorial about the Project. Computer Graphics Some Tutoral about the Project Lecture 6 Rastersaton, Antalasng, Texture Mappng, I have already covered all the topcs needed to fnsh the 1 st practcal Today, I wll brefly explan how to start workng on

More information

Modeling, Manipulating, and Visualizing Continuous Volumetric Data: A Novel Spline-based Approach

Modeling, Manipulating, and Visualizing Continuous Volumetric Data: A Novel Spline-based Approach Modelng, Manpulatng, and Vsualzng Contnuous Volumetrc Data: A Novel Splne-based Approach Jng Hua Center for Vsual Computng, Department of Computer Scence SUNY at Stony Brook Talk Outlne Introducton and

More information

Vol. 5, No. 3 March 2014 ISSN Journal of Emerging Trends in Computing and Information Sciences CIS Journal. All rights reserved.

Vol. 5, No. 3 March 2014 ISSN Journal of Emerging Trends in Computing and Information Sciences CIS Journal. All rights reserved. Journal of Emergng Trends n Computng and Informaton Scences 009-03 CIS Journal. All rghts reserved. http://www.csjournal.org Unhealthy Detecton n Lvestock Texture Images usng Subsampled Contourlet Transform

More information

CSCI 104 Sorting Algorithms. Mark Redekopp David Kempe

CSCI 104 Sorting Algorithms. Mark Redekopp David Kempe CSCI 104 Sortng Algorthms Mark Redekopp Davd Kempe Algorthm Effcency SORTING 2 Sortng If we have an unordered lst, sequental search becomes our only choce If we wll perform a lot of searches t may be benefcal

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

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

Image Segmentation with Perceptual Guidance

Image Segmentation with Perceptual Guidance Image Segmentaton wth Perceptual Gudance Xaofeng M Department of Computer Scence Rutgers Unversty xm@cs.rutgers.edu Fgure 1. Left top: the orgnal Davd head mage, left bottom, segmented mage wth mean shft

More information

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

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

More information

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

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 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

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

Type-2 Fuzzy Non-uniform Rational B-spline Model with Type-2 Fuzzy Data

Type-2 Fuzzy Non-uniform Rational B-spline Model with Type-2 Fuzzy Data Malaysan Journal of Mathematcal Scences 11(S) Aprl : 35 46 (2017) Specal Issue: The 2nd Internatonal Conference and Workshop on Mathematcal Analyss (ICWOMA 2016) MALAYSIAN JOURNAL OF MATHEMATICAL SCIENCES

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

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

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

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

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

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

More information

Joint Example-based Depth Map Super-Resolution

Joint Example-based Depth Map Super-Resolution Jont Example-based Depth Map Super-Resoluton Yanje L 1, Tanfan Xue,3, Lfeng Sun 1, Janzhuang Lu,3,4 1 Informaton Scence and Technology Department, Tsnghua Unversty, Bejng, Chna Department of Informaton

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

Maximum Variance Combined with Adaptive Genetic Algorithm for Infrared Image Segmentation

Maximum Variance Combined with Adaptive Genetic Algorithm for Infrared Image Segmentation Internatonal Conference on Logstcs Engneerng, Management and Computer Scence (LEMCS 5) Maxmum Varance Combned wth Adaptve Genetc Algorthm for Infrared Image Segmentaton Huxuan Fu College of Automaton Harbn

More information

Similarity-based denoising of point-sampled surfaces *

Similarity-based denoising of point-sampled surfaces * Wang et al. / J Zhejang Unv Sc A 008 9(6):807-85 807 Journal of Zhejang Unversty SCIENCE A ISSN 673-565X (Prnt); ISSN 86-775 (Onlne) www.zju.edu.cn/jzus; www.sprngerlnk.com E-mal: jzus@zju.edu.cn Smlarty-based

More information

An Improved Image Segmentation Algorithm Based on the Otsu Method

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

More information

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

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

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

Reading. 14. Subdivision curves. Recommended:

Reading. 14. Subdivision curves. Recommended: eadng ecommended: Stollntz, Deose, and Salesn. Wavelets for Computer Graphcs: heory and Applcatons, 996, secton 6.-6., A.5. 4. Subdvson curves Note: there s an error n Stollntz, et al., secton A.5. Equaton

More information

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

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

More information

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

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints Australan Journal of Basc and Appled Scences, 2(4): 1204-1208, 2008 ISSN 1991-8178 Sum of Lnear and Fractonal Multobjectve Programmng Problem under Fuzzy Rules Constrants 1 2 Sanjay Jan and Kalash Lachhwan

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

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

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

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

More information

3D vector computer graphics

3D vector computer graphics 3D vector computer graphcs Paolo Varagnolo: freelance engneer Padova Aprl 2016 Prvate Practce ----------------------------------- 1. Introducton Vector 3D model representaton n computer graphcs requres

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

Interpolation of the Irregular Curve Network of Ship Hull Form Using Subdivision Surfaces

Interpolation of the Irregular Curve Network of Ship Hull Form Using Subdivision Surfaces 7 Interpolaton of the Irregular Curve Network of Shp Hull Form Usng Subdvson Surfaces Kyu-Yeul Lee, Doo-Yeoun Cho and Tae-Wan Km Seoul Natonal Unversty, kylee@snu.ac.kr,whendus@snu.ac.kr,taewan}@snu.ac.kr

More information

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

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

More information

Positive Semi-definite Programming Localization in Wireless Sensor Networks

Positive Semi-definite Programming Localization in Wireless Sensor Networks Postve Sem-defnte Programmng Localzaton n Wreless Sensor etworks Shengdong Xe 1,, Jn Wang, Aqun Hu 1, Yunl Gu, Jang Xu, 1 School of Informaton Scence and Engneerng, Southeast Unversty, 10096, anjng Computer

More information