Augmented Distinctive Features for Efficient Image Matching

Size: px
Start display at page:

Download "Augmented Distinctive Features for Efficient Image Matching"

Transcription

1 Augmented Dstnctve Features for Effcent Image Matchng Quan Wang, We Guan and Suya You CGIT/IMSC USC Los Angeles, CA and Abstract Fndng correspondng mage ponts s a challengng computer vson problem, especally for confusng scenes wth surfaces of low textures or repeated patterns. Despte the well-known challenges of extractng conceptually meanngful hgh-level matchng prmtves, many recent works descrbe hgh-level mage features such as edge groups, lnes and regons, whch are more dstnctve than tradtonal local appearance based features, to tackle such dffcult scenes. In ths paper, we propose a dfferent and more general approach, whch treats the mage matchng problem as a recognton problem of spatally related mage patch sets. We construct augmented sem-global descrptors (ordnal codes) based on subsets of scale and orentaton nvarant local keypont descrptors. Ted rankng problem of ordnal codes s handled by ncreasngly keypont samplng around mage patch sets. Fnally, smlartes of augmented features are measured usng Spearman correlaton coeffcent. Our proposed method s compatble wth a large range of exstng local mage descrptors. Expermental results based on standard benchmark datasets and SURF descrptors have demonstrated ts dstnctveness and effectveness. 1. Introducton and related works Correspondence s one of the fundamental and ntensvely studed problems n computer vson. Approaches based on texture analyss around local nterest ponts [1, 2, 3, 4, 5, 6] have ganed much attenton recently due to ther robustness to dstorton and occluson, and avodance of dffcult problems such as edge detecton and segmentaton. However, when used alone, the localty property of such features frequently leads to nosy results. Geometrc constrants and consstency check generally need to be appled as addtonal layer to refne the result. Furthermore, although expermentally proven to be remarkably robust and nvarant to vewpont changes and dstortons usually at the cost of heavly computatonal demands, local features are also typcally not very dstnctve bascally because of the lack of bult-n geometry nformaton. Therefore, they tend to face great dffculty for scenes wthout hgh-textured surfaces or wth repeated patterns, easly confuse texture-based methods. To tackle the problem, many recent works n matchng and recognton domans try to drectly extract and descrbe hgh-level matchng prmtves such as constellaton of edges [7, 8] and regons [9, 10, 25] to enhance the descrptors dstnctveness. However, the accurate and clean acquston of those prmtves remans a challengng open problem n computer vson. The acqured curves and shapes are often nsuffcent stable between mages due to vewpont and lghtng changes. Input Images keypont detecton Sub- Images offset aggregaton keypont descrpton Relatve Features compute ranks Ordnal Codes handle ted Augmented Features Database Fgure 1: Overvew of our feature augmentaton process Ths paper presents our novel approach n a general framework to generate more dstnctve features based on nvarant local feature descrptors, wthout the need of hgh-level feature extracton. We propose to treat the classc mage matchng problem as a collecton of mage recognton problems, whch naturally ntegrates geometry nformaton nto our augmented features. Fg. 1 shows the maor steps nvolved when generatng augmented features. Frst, we detect stable nterest ponts n nput mages and extract correspondng mage patches based on postons, orentatons and scales of those ponts. Sub-mages, whch are bascally subsets of mage patches that are close to each other n mage space are constructed, consequently convertng the orgnal problem nto an mage recognton problem for those sub-mages. Next, the sem-global descrptor for each sub-mage s computed by frst accumulatng offsets between the sub-mage center and member patches descrptors, whch produces what we 1

2 called relatve features, contanng geometry nformaton of the neghborhood. The augmented features (ordnal codes) are generated by consderng the relatve rankngs of features components nstead of ther orgnal values, whch can be regarded as a feature normalzaton process. From certan pont of vew, our feature augmentaton process s analogous to the codng process n the general framework of mage recognton. One dfference s what we are codng here are sub-mages, not local patches (e.g. [19]) or the whole mage (e.g. [24]). Addtonally, we propose ncreasngly keypont samplng around submages to handle ted rankng problem of ordnal codes. Fnally, Spearman correlaton s used to measure the smlarty between augmented features and establsh the pont-to-pont correspondences for mage pars. Extensve expermental results based on SURF features have demonstrated the effectveness of our augmented features. We conduct experences usng benchmark datasets plus precson-recall analyss, together wth supplemental real-word challengng mage pars, both ndcatng hgher dstnctveness and lower outler level of our augmented features compared wth base features. The addtonal computatonal cost of the feature augmentaton process s nomnal, whch makes buldng nteractve or even real-tme computer vson applcatons possble when combned wth fast base features. Furthermore, our feature augmentaton framework s general enough to be compatble wth a wde range of exstng nvarant local features, provdng mportant performance gan to mage representaton methods or upgradng exstng mage feature databases at a mnmum cost. We also provde theoretcal analyss about compatblty ssues of our method workng wth dfferent knds of local features. Our proposed approach s related to the followng recent works: In [11], Boureau, et al. provded a systematc evaluaton about combnatons of varous codng and poolng technques for recognton, and concluded that large performance ncrease can be obtaned by merely representng neghborng descrptors ontly. Smlar dea was explored n [12] whch also proposed a novel way to optmze dmenson reducton and approxmated nearest neghbor searchng at the same tme. Concernng the ordnal descrpton method we used, the early work of ordnal measurement tracks back to M. Kendall [15]. Recent works applyng ordnal descrpton on mage correspondence and recognton problems nclude R. Zabh and J. Woodfll [16], J. Luo, et al. [17] and M. Toews and W. Wells [18], whch bult upon SIFT descrptors and reported superor results n terms of precson-recall when compared wth many wdely-used descrptors such as orgnal SIFT, PCA-SIFT and GLOH. Our sub-mage concept s also enlghtened by the followng classc technques: frst of all, the shape context [20] takes sampled contour ponts as nputs and constructs shape descrptors n log-polar hstograms usng relatve postons (drecton and dstance) between contour ponts. Two fundamental dfferences are: our sub-mages are descrbed by countng offsets n nvarant feature space, not n mage space. And the constructon of sub-mages s drectly based on stable nterest ponts, therefore avodng the need to acqure clean shapes and contour ponts from mages, whch s a challengng fundamental vson problem by tself. Second, our sub-mage descrpton s also related to Fsher kernel, whch provdes the offset drectons n parameter space nto whch the learnt dstrbuton should be modfed n order to better ft the newly observed data. Recent work of Perronnn et al. [21] uses fsher kernel to obtan compact and effcent mage representatons for mage classfcaton. Last, the famous bag-of-features approach [22] uses K-mean to generate code-books based on nvarant local features, assgns each query descrptor to one tem n the code-book, producng a hstogram of codes representng the whole mage, whle our approach proposes sub-mage as matchng prmtves and uses accumulated offsets nstead of the cluster centers n the feature space. Each base descrptor s converted nto exactly one augmented descrptor, facltatng dense pontto-pont correspondences. 2. Feature augmentaton process Ths secton presents the reasonng and detals of our feature augmentaton process. Based on keypont locatons and descrptons, we construct sub-mages representng nterested neghborhood n mage space, then compute relatve features for each sub-mage ntegratng geometry nformaton n feature space, whch are normalzed usng ordnal descrpton and produce fnal augmented features Sub-mages Many exstng mage matchng methods use local mage patches around nterested keyponts as matchng prmtves. The mage patches can be extracted ether from scale and rotaton-adaptve neghborhoods, where transformaton parameters are determned through searchng n scale space and orentaton hstograms (e.g. [2], [3], [4] and [5]), or from regular regons of fxed szes, whch acheve vewpont nvarance through separated multple-vew tranng process (e.g. [6], [19] and [23]). The produced local features can be very robust aganst vewpont changes and dstortons, but usually nsuffcent dstnctve due to the lack of global geometry nformaton. On the other hand, mage classfcaton and recognton works focus more on dstnctveness among dfferent mages rather than robustness, generally producng one unfed descrptor for each mage and can by no means provde robust pont-to-pont correspondences. We propose to use sub-mages as matchng prmtves, amng to fll the gap between the above two extremes. Sub-mages are sets of local mage patches that are close 2

3 to each other n mage space (Fg. 2). As relatve local structures, they are robust to factors such dstortons and occlusons. They also ntegrate sem-global geometry nformaton n order to mprove feature dstnctveness. Fgure 2: Illustratons of sub-mage concept (k = 5 cases). Red and blue dots are leaders and members of each sub-mage respectvely. Grey crosses are other nterest ponts. To construct sub-mages, we frst detect stable nterest ponts (represented by P ) and extract fxed or nvarantly adaptve mage patches around each nterest pont dependng on whch detector s used. After the poston of each P s stored and properly ndexed (herarchy tree structure preferred especally when the total number of keyponts, represented as num, s large), we construct one sub-mage structure for each P (represented by S(P )) wth P as ts leader and ts k nearest neghbors n mage space as members. S( P) = { P} { k NN( P)}, 1 num Each S(P ) s an abstract type of mage contanng nformaton for the neghborhood of P. Both leaders and members of sub-mages are centrally organzed by keyponts. As a result, the keypont detector should be suffcent stable so that the consttuton of the same submage wll reman consstent to certan degree between dfferent nput mages under varous vewng condtons. In our current experments, we use the effcent keypont detector n [3] based on ntegral mages, whch s very fast to apply and provdes us wth suffcent repeatablty. After the sub-mage constructon, fndng matchng ponts n the nput mage par s equvalent to classfyng a large set of sub-mages generated from both mages Relatve features Gven sub-mages, our next task s to form dstnctve descrptors, whch wll be used later to measure the smlartes between sub-mages effcently. Ths step s analogous to the codng process n mage recognton. Suppose we choose an nvarant local feature descrptor (wth descrptor dmensonalty represented by dm) and D(P ) s the feature vector of P and D (P ) represents ts - th vector component. Notce that there s no need to adopt the keypont detector and descrptor all belongng to the same mage matchng system, as long as the detector selected s stable and the descrptor nvarant, meetng general requrement of those components. Ths provdes extra flexblty for our augmented feature to work well wth varous combnatons of exstng keypont detecton and mage representaton technques. If we seek to combne the sub-mage s descrptors drectly, for example, smply concatenatng D (P ) for all the P belongng to the same sub-mage, at least two drawbacks wll mmedately follow. One s the rather hgh dmensonalty of the resultng descrptors (when k = 5 and standard SURF [3] descrptor s used, each sub-mage wll be assocated wth a 384-D vector), whch rases practcal obstacles to later ndexng and searchng steps due to dmensonalty curse. The other s such combnaton wll hghly lkely compromse the nvarant propertes of the base descrptors because of the dfferent orders when performng the concatenaton. Encouraged by successful classc technques such as Fsher kernel [21] and shape context [22], we beleve that generally speakng, features based on relatve and aggregated values (ether n mage space lke shape context or n feature space lke macro features n [11]) usually demonstrate more robust performance than features drectly based on raw values of base descrptors. Fgure 3: Sub-mages (red and blue crosses) of smlar physcal locatons n one mage matchng par, and ther correspondng relatve features vsualzed as 2D (8 by 8) hstogram patterns. Based on the above reasonng and analyss, we compute relatve features for S(P ) (denoted as R(P )) by accumulatng k offsets between leader and member descrptors of S(P ) for each dmenson. Formally, we can defne: Then the -th dmenson of the relatve feature s computed as: 3

4 R ( P ) = δ n [1, num], n D ( P ) D ( Pn ), 1 dm The computed relatve features have a constant low dmensonalty regardless of k, the number of members n each sub-mage. Moreover, assumng the bases descrptors are scale and rotaton-nvarant, and the keypont detector s stable, t can be easly proved that the relatve features produced wll also be nvarant, snce the sub-mage members are consdered n an order-ndependent way. Ths s a consderable advantage for mage matchng applcatons, acheved by computng relatve features n nvarant feature space and mantanng nvarance durng the computaton process. As a comparson, features computed by aggregatng offsets n mage space, such as shape context, are fundamentally not nvarant to orentatons and typcally obtan lmted degree of scale nvarance through normalzaton technques. In our current experments, we use SURF descrptor [3] wth dm = 64. We vsualze the generated relatve features by frst transformng ther components nto 0~255 ntensty range and then renderng each descrptor as a 2D (8 by 8) hstogram. Fg. 3 shows some of our relatve features generated usng a modfed verson of the famous testng mage par from [2] (extra lghtng changes and dstortons are added) Normalzaton After relatve features n each mage are computed, we can drectly measure ther smlartes usng standard dstance metrcs such as Eucldean dstance and establsh the ntal correspondences between two feature sets. For smple mage matchng pars, e.g. one mage s the n-plane rotated or unformly scaled verson of the other, accurate results could be expected. However, when t comes to challengng mage pars nvolvng complcated vewponts and lghtng changes (e.g. Fg. 3), we found such ntal matchng results are generally very nosy. The prmary reason les n the lnear relatonshp assumpton of today s wdely-used computatonal effcent metrcs such as Eucldean or ch-squared dstances. To handle the volaton of lnear assumpton (e.g. from llumnaton changes), orgnal feature vectors are usually normalzed to remove non-lnearty before enterng the matchng process. Many exstng methods use a set of experences- or experments-determned parameters to threshold, translate and rescale, fttng the orgnal data nto a fxed range, whch s naturally dffcult to generalze to other vewng condtons or combnatons of stuatons. As can be observed from fgure 3, n terms of raw values magntudes (the ntenstes of the 2D hstogram cells), sub-mages of smlar physcal locatons could produce rather dfferent relatve features due to vewng condton changes, whch s exactly the reason of nosy results when drectly measurng the smlartes of relatve features. However, t can also be observed that although those bns absolute values are not dentcal across dfferent vews, ther relatve orders are overall consstent. Based on the above analyss and observatons, we propose to use ordnal descrpton as a parameter-free and computatonal effcent approach to normalze our relatve features. Frst, for each relatve feature, we locally sort ts vector components and produce a sorted array. The augmented feature (represented by A(P )) s generated by replacng the orgnal values wth ts correspondng ranks n the sorted array for every dmenson. The overall computatonal complexty of normalzng each relatve feature s only dm*log(dm) due to sortng. A ( P ) = { R ( P ) R ( P ) R ( P )} 1 dm m m Our fnal augmented features consder the relatve rankng of features each dmenson nstead of ther orgnal values. As a result, each augmented feature s represented as an nteger vector of dm dmensons, more specfcally, a permutaton of the set {1, 2,, dm}. The ordnal descrpton we use generates normalzed mage representatons nvarant to monotonc deformatons and also robust aganst certan degree of challengng nonlnear and non-unform factors such as partal lghtngs. Snce each ordnal codes s a permutaton of nteger set {1, 2,, dm}, we treat those ntegers as ntensty values and smply vsualze each ordnal codes as a barcodes-lke 1D pattern. Fgure 4 shows the barcodes assocated wth the same sub-mages as n fg.3 n the two nput mages. It ndcates that after the ordnal descrpton, our feature smlarty s notably mproved. Under dfferent vewng condtons, features belongs to the same physcal locatons are more smlar whle others reman dstnctve. Fgure 4: Augmented features vsualzaton for smlar physcal locatons n the two nput mages. The left column s extracted from the left mage whle the rght column from the rght mage. One crucal ssue for ordnal descrpton s how to handle ted rankng. Gven a relatve feature vector, when ts two components wth dfferent ndces contan the same orgnal value, the produced ordnal codes wll contan ted dentcal rankngs for the two ndces, resultng n 4

5 unexpected outputs. The smplest way to handle ted rankng s to use the dfferent ndces as fal safe reference to break the ted stuaton. Ths method s equvalent to usng a vector of orgnal ndces as a reference. In [18], such reference vector s obtaned by averagng a large set of descrptors, e.g. the descrptors of the whole mage, offlne, whch s approprate for codng the whole mage for recognton purpose. However, we argue that for mage matchng, the reference vector used by sem-local ordnal descrpton should reflect the local orderng trend, whch s usually not consstent wth the global trend. Algorthm: Ted Rankng Breaker 1: Assume: R = R ; ; 2: Intalze: set S = {all keyponts}-s(p ); 3: do teraton t = 1, 2,, M; 4: P n = NN(P, S); 5: S = S P n ; 6: S(P ) = S(P ) + P n ; 7: compute new R(P, S(P )); 8: f (R R ) break; 9: end teraton; 10: f (R R ) use (R, R ) to handle ted rankng; 11: else use (, ) nstead; Therefore, we propose ncreasng keypont samplng around sub-mages to break ted rankng. The dea s to teratvely nclude addtonal nearest neghbors of the current sub-mage one by one and compute new relatve features durng each teraton untl the ted rankng s broken, or, for effcency consderaton, untl a maxmum number of teraton (denoted by M) s reached, n whch case, we use the orgnal ndces as reference vector nstead. Pseudo codes for ted rankng handlng are provded above. NN(P,S) s a functon returnng the nearest neghbor of P n S. 3. Establshng correspondences Although standard dstance metrcs e.g. Eucldean dstance can be used to measure the smlartes, better comparson results can be obtaned more effcently usng specal metrcs based on characterstcs of the ordnal codes, for example, gven dm, there are a fxed total number of unque codes (dm!) and are all of the same length. Ordnal descrpton and measurement have been studed for nearly a century and many specally desgned dstance metrcs were proposed. In ths paper, we studed three measurements compatble well wth our features and problem doman. Frst s the measurement of element-wse consstency countng. Assume A and A are two augmented feature vectors for two sub-mages, and A s the -th vector components. Ther smlarty can be effcently computed as: Dst( A, A' ) = { A A A' } 1 dm The second metrc s based on relatve orderng of vector component pars measured by sgn() functon, whch returns 1 f the nput parameters have the same sgn and -1 otherwse. Formally, the Kendall coeffcent [15] s defned as the followng: Dst ( A, A' ) = 2 [ 1,dm] m [ + 1, dm] sgn( A A m dm(dm 1), A' A' Last, based on the ordnal codes property of equal length, our modfed verson Spearman correlaton coeffcent [14] returns the smlarty of two ordnal codes n the range of 0~ ( A A' ) [1,dm] Dst ( A, A' ) = 1 2 dm(dm 1) Our own experments ndcate that the Spearman correlaton coeffcent generally provdes better matchng results than the other two. As addtonal but rather practcal consderatons, ts conventonal unt range also works well wth many exstng mage matchng components and frameworks, such as dstance rato (used n [2] and [5]) and consstent check methods lke RanSac. Therefore, we select the Spearman correlaton coeffcent as our dstance metrc n the next secton s evaluaton. 4. Expermental results Our proposed augmented feature has been ntensvely tested usng standard benchmark datasets [5] wth known ground truth. We prmarly focus on outdoor scenes accordng to our proect s nterest. Besdes the standard test data, we also conducted experments on other realworld mage pars of outdoor nature scenes wth manmade obects such as buldngs and sgns. Many of those addtonal testng mage pars contan nterest obects wth surfaces of low-textures or repeated patterns, partcularly challengng for varous exstng local mage features. We experment dfferent knds of nvarant mage descrpton methods and fnally select SURF (Speeded Up Robust Feature) [3] as our base descrptor, because t provdes the best tradeoff between matchng speed and feature robustness. SURF s bult upon other expermentsverfed and successful detectors and descrptors, such as famous SIFT [2], but smplfy those steps to the essental. It proposes the use of ntegral mages to drastcally reduce the number of operatons for smple box convolutons, ndependent of the chosen scale. The feature descrpton s based on sums of Haar wavelet components, whch can be constructed and matched more effcently compared wth other state-of-the-art methods. In our evaluaton, the wdely-used 64 dmenson of SURF descrptor s used. We set k, the only parameter of the whole feature m ) 5

6 augmentaton process, to 5 whch we expermentally fnd to provde best results. Overall, our expermental results demonstrate that compared wth base descrptors, the proposed augmented features acheve remarkably hgher level of dstnctveness wthout loss of robustness, at a nomnal addtonal computatonal cost. Specfcally, besdes outperformng base features n the standard precson-recall curves, the new augmented feature ntegratng geometry nformaton generally produces larger number of correspondences under the same dstance rato requrement and less number of nosy matchngs especally for challengng real-world scenes. The proposed method s ntegrated nto our mage matchng system, whch generates augmented features of one target mage offlne then matches the features of query mages onlne and frame by frame. The total processng tme each frame ncludng keypont detecton, descrpton, feature augmentaton process and fnally establshng correspondences takes only 100~150ms wth a peak memory occupaton of around 15M, makng the system an attractve soluton for mage matchng problems of nteractve/real-tme applcatons such as outdoor navgaton and augmented realty on small moble devces Standard tests The standard test sets we expermented consst of mage sets of varous scenes (K. Mkolaczyk and C. Schmd [5]), wth each contanng sx mages (fve pars when the frst mage s fxed as reference mage) of successve ncrements of one type of mage deformaton ncludng mage blur and compresson, lghtng and vewpont changes. One mage transformaton matrx s assocated wth each mage par, so that for any keypont poston n the reference mage, we are able to compute the ground truth correspondng poston n any of the other fve mages, verfyng matchngs returned by varous methods. Durng the evaluaton process, for each mage par, we construct base descrptors and two versons of augmented descrptors usng local (L) and global (G) reference vector respectvely, for the same set of nterest ponts and then use Eucldean dstance and Spearman correlaton coeffcent to measure ther smlartes and establsh ntal matchngs. Wthn one test, the same dstance rato s appled to both base and augmented tracks to obtan fnal reported matchngs. Dstance ratos are vared n dfferent tests to generate the fgure data ponts. Next, for each keypont poston n the reference mage, by comparng ts reported matchng postons wth ground truth matchng postons, we are able to dstngush true and false matchngs. Ths crucal nformaton, combned wth the total number of reported and possble correspondences, allow us to compute precson and recall for one mage par, under one dstance rato. Accordng to our proect nterests, we focus our experments on challengng outdoor scenes wth nonplanar obects and low-textured surfaces, handlng complcated vewpont changes. To generate fgure 5, we choose 13 mage pars meetng the above gudelnes from three mage sets (the boat set, the grafft set, and the bke set), whch also represents challengng factors n outdoor moble applcatons such as rotatons of the devce and blurred nput mages. Each mage par s tested by usng base and augmented features and ten dfferent dstance ratos. Fgure 5 shows the precson and recall results summary of over 300 mage matchng tests, demonstratng the superor performance of our proposed method. Recall Precson Aug. Feature (L) Aug. Feature (G) Base Feature Fgure 5: Recall-precson for our standard test results summary, evaluated on 64-D SURF as base descrptors. Results averaged over 13 mages pars of K. Mkolaczyk s testng data set Dense matchng from dstnctve features In ths subsecton, we demonstrate that our augmented features are more dstnctve than base features, resultng denser and cleaner correspondences. Towards ths goal, we compute base and augmented features usng the fve mage pars of boat mage set (wth ncreasng level of zoom and rotaton), apply the same dstance rato and record the number of reported matchngs for the two feature tracks respectvely. Snce dstance rato s defned as the rato of the best matchng s smlarty over the second best, we beleve that from some aspects of vew, the number of remanng matchngs after applyng a fxed dstance rato can also be used as a smple ndcator of feature dstnctveness. Assume one method generates smlar, clustered thus less dstnctve features whle the other produces features well scattered and dstrbuted n the feature space. After applyng the same dstance rato flter, the frst method wll have much less matchngs remanng because the dfference between ts dfferent matchng s smlartes s much lmted. 6

7 350 Total Number of Features # of Reported Matchngs (Aug.) # of Reported Matchngs (Base) representatons such as [13] and [23] couldn t beneft from the feature augmentaton process. Number of Features and Matchngs Test Index (Increasng Zoom and Rotaton) Fgure 6: Number of matchngs remanng after applyng the same dstance rato. Evaluaton based on the boat mage sets. Fgure 6 shows that our augmented features lead to sgnfcant larger number of reported matchngs than the base features, under the same dstance rato. Ths result can also be consdered ontly wth secton 4.1, whch ndcates better precson and recall tradeoff of the denser matchngs, together wth secton 4.3, whch demonstrates consstent results for mages outsde the standard datasets as well Matchng vsualzaton and analyss Ths subsecton provdes some lne by lne matchng vsualzaton results for both standard test mages and other real-world scenes. We apply the same dstance rato and for each of remanng fnal matchng, draw a blue lne from ts keypont n reference mage to the matched keypont n the current query mage. As can be observed from fgure 7 and consstent wth the quanttatve evaluaton results from the above two subsectons, through augmented features, we are able to obtan denser correspondences and less matchng outlers than usng base features. The prmary reason les n the fact that our augmented features enhance dstnctveness by ntegratng sem-global geometry nformaton based on sub-mages concepts and offset aggregaton n feature space, whle mantanng and even mprovng feature nvarance and robustness through ordnal descrpton. Last, regardng the compatblty ssue of our feature augmentaton, theoretcally the proposed method should work well wth a large range of exstng nvarant local features ncludng but not lmt to SIFT [2], SURF [3], and GLOH [5]. However, there are a few ssues worth consderng beforehand. Frst of all, t goes wthout sayng that the base feature needs to have descrptors n vector form. Otherwse the computaton of relatve features and ordnal descrpton can not be appled drectly. So specal approaches based on drect learnng on local mage patches wthout vector Fgure 7: Matchng vsualzaton and comparson of base (top) and augmented (bottom) features. Standard dataset: (a) Grafft; (b) Boat; Extra testng dataset: (c) buldng wth smlar or repeated patterns; (d) street scene wth large vewpont changes. Next, bag-of-features [22] and some mult-vew based mage matchng technques such as [6] and [19] use K- mean cluster centers nstead of the orgnal nterest ponts to represent mages or features. In such case, specal cautons are needed to make sure except stable keyponts, those cluster centers are also suffcent stable otherwse t s hard to guarantee that sub-mages, as the very foundaton of our feature augmentaton process, wll 7

8 reman repeated and consstent across dfferent vewng condtons. Fnally, concernng the normalzaton component n our method, ordnal descrpton generally works better when the feature vectors t apples to have a hgh dmensonalty, provdng a large pool of unque codes. Image matchng methods advocatng dmenson reducton (e.g. by usng PCA) of ther feature vectors, such as [4] and [19] are ll-suted for our normalzaton component. It s reported n [18] that drectly applyng ordnal descrpton onto PCA-SIFT [4] can even lead to nferor performance. 5. Concluson Ths paper presents our feature augmentaton process, producng more dstnctve features for effcently mage matchng. The proposed method s bult upon the concept of sub-mages, whch connects close small mage patches n mage space, convertng one mage matchng problem nto a collecton of mage recognton problems. Our relatve features aggregate descrptor offsets n nvarant feature space and wthn sub-mages, n order to ntegrate geometry nformaton and produce sem-global features. Based on vsualzaton and analyss of correspondng relatve features, we propose to use ordnal descrpton to normalze and generate augmented features, nvarant to monotonc deformatons and beyond. Fnally, smlartes are measured by Spearman correlaton coeffcent and correspondences are establshed. Expermental results usng standard and supplemental datasets verfed the augmented features contanng addtonal geometry nformaton are more dstnctve, leadng to denser and cleaner correspondences. References [1] C. Schmd and R. Mohr, Local grayvalue nvarants for mage retreval, IEEE Transactons on Pattern Analyss and Machne Intellgence, pp , May [2] D. Lowe. Dstnctve mage features from scale-nvarant keyponts. Internatonal Journal of Computer Vson, [3] H. Bay, T. Tuytelaars and L. V. Gool, SURF: Speeded Up Robust Features, Proceedngs of the nnth European Conference on Computer Vson, May [4] Y. Ke and R. Sukthankar. PCA-SIFT: A more dstnctve representaton for local mage descrptors. Computer Vson and Pattern Recogn-ton, pages , [5] K. Mkolaczyk, C. Schmd, A performance evaluaton of local descrptors. In PAMI 27(10): , [6] Q. Wang and S. You, Real-Tme Image Matchng Based on Multple Vew Kernel Proecton, IEEE Conference on Computer Vson and Pattern Recognton, June [7] O. Danelsson, S. Carlsson and J. Sullvan, Automatc learnng and extracton of mult-local features, 2009 IEEE 12th Internatonal Conference on Computer Vson, pp , Sept Oct [8] Z. S, H. Gong, Y. Wu and S. Zhu, Learnng mxed templates for obect recognton, Computer Vson and Pattern Recognton, pp , June [9] Q. Wang and S. You, A Vson-based 2D-3D Regstraton System, IEEE Workshop on Applcatons of Computer Vson (WACV), Snowbrd, Utah, December 7-8, 2009 [10] C. Gu, J.J. Lm, P. Arbelaez and J. Malk, Recognton usng regons, IEEE Conference on Computer Vson and Pattern Recognton, pp , June [11] Y.L. Boureau, F. Bach, Y. LeCun and J. Ponce, Learnng md-level features for recognton, 2010 IEEE Conference on Computer Vson and Pattern Recognton, pp , June [12] H. Jegou, M. Douze, C. Schmd and P. Perez, Aggregatng local descrptors nto a compact mage representaton, 2010 IEEE Conference on Computer Vson and Pattern Recognton, pp , June [13] M. Calonder, V. Lepett, P. Fua, K. Konolge, J. Bowmanand P. Mhelch, Compact sgnatures for hghspeed nterest pont descrpton and matchng, 2009 IEEE 12th Internatonal Conference on Computer Vson, pp , Sept Oct [14] C. E. Spearman. Proof and measurement of assocaton between two thngs. Amercan Journal of Psychology, 15:72 101, [15] M. Kendall. A new measure of rank correlaton. Bometrka, 30:81 89, [16] R. Zabh and J. Woodfll, Non-parametrc local transforms for computng vsual correspondance. European Conference on Computer Vson, pages , [17] J. Luo, S. P. Etz, and R. T. Gray. Normalzed Kemeny and Snell Dstance: A Novel Metrc for Quanttatve Evaluaton of Rank-Order Smlarty of Images. IEEE TPAMI, 24(8): , [18] M. Toews and W. Wells, SIFT-Rank: Ordnal descrpton for nvarant feature correspondence, 2009 IEEE Conference on Computer Vson and Pattern Recognton, pp , June [19] V. Lepett, J. Plet and P. Fua, Pont matchng as a classfcaton problem for fast and robust obect pose estmaton, IEEE Computer Socety Conference on Computer Vson and Pattern Recognton, [20] S. Belonge, J. Malk and J. Puzcha, Shape matchng and obect recognton usng shape contexts, IEEE Transactons on Pattern Analyss and Machne Intellgence, vol.24, no.4, pp , Apr [21] F. Perronnn and C. R. Dance. Fsher kernels on vsual vocabulares for mage categorzaton. IEEE Conference on Computer Vson and Pattern Recognton, June [22] J. Svc and A. Zsserman. Vdeo Google: A text retreval approach to obect matchng n vdeos. IEEE Internatonal Conference on Computer Vson, [23] V. Lepett and P. Fua, Keypont Recognton usng Randomzed Trees, In PAMI, Vol. 28, Nr. 9, pp , [24] S. Lazebnk, C. Schmd, and J. Ponce, Beyond Bags of Features: Spatal Pyramd Matchng for Recognzng Natural Scene Categores, IEEE Conference on Computer Vson and Pattern Recognton, [25] J. Km and K. Grauman, Asymmetrc Regon-to-Image Matchng for Comparng Images wth Generc Obect Categores, Computer Vson and Pattern Recognton, pp , June

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

Feature Reduction and Selection

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

More information

A 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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

Improved SIFT-Features Matching for Object Recognition

Improved SIFT-Features Matching for Object Recognition Improved SIFT-Features Matchng for Obect Recognton Fara Alhwarn, Chao Wang, Danela Rstć-Durrant, Axel Gräser Insttute of Automaton, Unversty of Bremen, FB / NW Otto-Hahn-Allee D-8359 Bremen Emals: {alhwarn,wang,rstc,ag}@at.un-bremen.de

More information

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

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

More information

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

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

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

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

More information

Classifier Selection Based on Data Complexity Measures *

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

More information

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

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

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

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

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

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

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

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

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

IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH

IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH Jyot Joglekar a, *, Shrsh S. Gedam b a CSRE, IIT Bombay, Doctoral Student, Mumba, Inda jyotj@tb.ac.n b Centre of Studes n Resources Engneerng,

More information

Real-time Motion Capture System Using One Video Camera Based on Color and Edge Distribution

Real-time Motion Capture System Using One Video Camera Based on Color and Edge Distribution Real-tme Moton Capture System Usng One Vdeo Camera Based on Color and Edge Dstrbuton YOSHIAKI AKAZAWA, YOSHIHIRO OKADA, AND KOICHI NIIJIMA Graduate School of Informaton Scence and Electrcal Engneerng,

More information

Image Matching Algorithm based on Feature-point and DAISY Descriptor

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

More information

TN348: Openlab Module - Colocalization

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

More information

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

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

Hierarchical clustering for gene expression data analysis

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

More information

LECTURE : 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

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

Optimizing Document Scoring for Query Retrieval

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

More information

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

Range Data Registration Using Photometric Features

Range Data Registration Using Photometric Features Range Data Regstraton Usng Photometrc Features Joon Kyu Seo, Gregory C. Sharp, and Sang Wook Lee Dept. of Meda Technology, Sogang Unversty, Seoul, Korea Dept. of Radaton Oncology, Massachusetts General

More information

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

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

More information

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

WIRELESS CAPSULE ENDOSCOPY IMAGE CLASSIFICATION BASED ON VECTOR SPARSE CODING.

WIRELESS CAPSULE ENDOSCOPY IMAGE CLASSIFICATION BASED ON VECTOR SPARSE CODING. WIRELESS CAPSULE ENDOSCOPY IMAGE CLASSIFICATION BASED ON VECTOR SPARSE CODING Tao Ma 1, Yuexan Zou 1 *, Zhqang Xang 1, Le L 1 and Y L 1 ADSPLAB/ELIP, School of ECE, Pekng Unversty, Shenzhen 518055, Chna

More information

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures

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

More information

A Clustering Algorithm for Key Frame Extraction Based on Density Peak

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

More information

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

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

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

CMPS 10 Introduction to Computer Science Lecture Notes

CMPS 10 Introduction to Computer Science Lecture Notes CPS 0 Introducton to Computer Scence Lecture Notes Chapter : Algorthm Desgn How should we present algorthms? Natural languages lke Englsh, Spansh, or French whch are rch n nterpretaton and meanng are not

More information

Multi-stable Perception. Necker Cube

Multi-stable Perception. Necker Cube Mult-stable Percepton Necker Cube Spnnng dancer lluson, Nobuuk Kaahara Fttng and Algnment Computer Vson Szelsk 6.1 James Has Acknowledgment: Man sldes from Derek Hoem, Lana Lazebnk, and Grauman&Lebe 2008

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

Histogram of Template for Pedestrian Detection

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

More information

A Gradient Difference based Technique for Video Text Detection

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

More information

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

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

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

More information

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

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

More information

A Gradient Difference based Technique for Video Text Detection

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

More information

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

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

Loop Transformations, Dependences, and Parallelization

Loop Transformations, Dependences, and Parallelization Loop Transformatons, Dependences, and Parallelzaton Announcements Mdterm s Frday from 3-4:15 n ths room Today Semester long project Data dependence recap Parallelsm and storage tradeoff Scalar expanson

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

Pictures at an Exhibition

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

More information

User Authentication Based On Behavioral Mouse Dynamics Biometrics

User Authentication Based On Behavioral Mouse Dynamics Biometrics User Authentcaton Based On Behavoral Mouse Dynamcs Bometrcs Chee-Hyung Yoon Danel Donghyun Km Department of Computer Scence Department of Computer Scence Stanford Unversty Stanford Unversty Stanford, CA

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

y and the total sum of

y and the total sum of Lnear regresson Testng for non-lnearty In analytcal chemstry, lnear regresson s commonly used n the constructon of calbraton functons requred for analytcal technques such as gas chromatography, atomc absorpton

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

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

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

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

Fitting and Alignment

Fitting and Alignment Fttng and Algnment Computer Vson Ja-Bn Huang, Vrgna Tech Many sldes from S. Lazebnk and D. Hoem Admnstratve Stuffs HW 1 Competton: Edge Detecton Submsson lnk HW 2 wll be posted tonght Due Oct 09 (Mon)

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

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

Querying by sketch geographical databases. Yu Han 1, a *

Querying by sketch geographical databases. Yu Han 1, a * 4th Internatonal Conference on Sensors, Measurement and Intellgent Materals (ICSMIM 2015) Queryng by sketch geographcal databases Yu Han 1, a * 1 Department of Basc Courses, Shenyang Insttute of Artllery,

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

BioTechnology. An Indian Journal FULL PAPER. Trade Science Inc.

BioTechnology. An Indian Journal FULL PAPER. Trade Science Inc. [Type text] [Type text] [Type text] ISSN : 0974-74 Volume 0 Issue BoTechnology 04 An Indan Journal FULL PAPER BTAIJ 0() 04 [684-689] Revew on Chna s sports ndustry fnancng market based on market -orented

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

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

Fast Feature Value Searching for Face Detection

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

More information

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

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

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

CSE 326: Data Structures Quicksort Comparison Sorting Bound

CSE 326: Data Structures Quicksort Comparison Sorting Bound CSE 326: Data Structures Qucksort Comparson Sortng Bound Steve Setz Wnter 2009 Qucksort Qucksort uses a dvde and conquer strategy, but does not requre the O(N) extra space that MergeSort does. Here s the

More information

Signature and Lexicon Pruning Techniques

Signature and Lexicon Pruning Techniques Sgnature and Lexcon Prunng Technques Srnvas Palla, Hansheng Le, Venu Govndaraju Centre for Unfed Bometrcs and Sensors Unversty at Buffalo {spalla2, hle, govnd}@cedar.buffalo.edu Abstract Handwrtten word

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

Detection of Human Actions from a Single Example

Detection of Human Actions from a Single Example Detecton of Human Actons from a Sngle Example Hae Jong Seo and Peyman Mlanfar Electrcal Engneerng Department Unversty of Calforna at Santa Cruz 1156 Hgh Street, Santa Cruz, CA, 95064 {rokaf,mlanfar}@soe.ucsc.edu

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

Visual Thesaurus for Color Image Retrieval using Self-Organizing Maps

Visual Thesaurus for Color Image Retrieval using Self-Organizing Maps Vsual Thesaurus for Color Image Retreval usng Self-Organzng Maps Chrstopher C. Yang and Mlo K. Yp Department of System Engneerng and Engneerng Management The Chnese Unversty of Hong Kong, Hong Kong ABSTRACT

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

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

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

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

Using the Visual Words based on Affine-SIFT Descriptors for Face Recognition

Using the Visual Words based on Affine-SIFT Descriptors for Face Recognition Usng the Vsual Words based on Affne-SIFT Descrptors for Face Recognton Yu-Shan Wu, Heng-Sung Lu, Gwo-Hwa Ju, Tng-We Lee, Yen-Ln Chu Busness Customer Solutons Lab., Chunghwa Telecommuncaton Laboratores

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