MLKD s Participation at the CLEF 2011 Photo Annotation and Concept-Based Retrieval Tasks

Size: px
Start display at page:

Download "MLKD s Participation at the CLEF 2011 Photo Annotation and Concept-Based Retrieval Tasks"

Transcription

1 MLKD s Partcpaton at the CLEF 2011 Photo Annotaton and Concept-Based Retreval Tasks Eleftheros Spyromtros-Xoufs, Konstantnos Sechds, Grgoros Tsoumakas, and Ioanns Vlahavas Dept of Informatcs Arstotle Unversty of Thessalonk Thessalonk 54124, Greece {espyrom,sechds,greg,vlahavas}@csd.auth.gr Abstract. We partcpated both n the photo annotaton and conceptbased retreval tasks of CLEF For the annotaton task we developed vsual, textual and mult-modal approaches usng mult-label learnng algorthms from the Mulan open source lbrary. For the vsual model we employed the ColorDescrptor software to extract vsual features from the mages usng 7 descrptors and 2 detectors. For each combnaton of descrptor and detector a mult-label model s bult usng the Bnary Relevance approach coupled wth Random Forests as the base classfer. For the textual models we used the boolean bag-of-words representaton, and appled stemmng, stop words removal, and feature selecton usng the ch-squared-max method. The mult-label learnng algorthm that yelded the best results n ths case was Ensemble of Classfer Chans usng Random Forests as base classfer. Our mult-modal approach was based on a herarchcal late-fuson scheme. For the concept based retreval task we developed two dfferent approaches. The frst one s based on the concept relevance scores produced by the system we developed for the annotaton task. It s a manual approach, because for each topc we manually selected the topcs and manually set the strength of ther contrbuton to the fnal rankng produced by a general formula that combnes topc relevance scores. The second approach s based solely on the sample mages provded for each query and s therefore fully automated. In ths approach only the textual nformaton was used n a query-by-example framework. 1 Introducton ImageCLEF s the cross-language mage retreval track run annually snce 2003 as part of the Cross Language Evaluaton Forum (CLEF) 1. Ths paper documents the partcpaton of the Machne Learnng and Knowledge Dscovery (MLKD) group of the Department of Informatcs of the Arstotle Unversty of Thessalonk at the photo annotaton task (also called vsual concept detecton and annotaton task) of ImageCLEF

2 Ths year, the photo annotaton task conssted of two subtasks. An annotaton task, smlar to that of ImageCLEF 2010, and a new concept-based retreval task. Data for both tasks come from the MIRFLICKR-1M mage dataset [1], whch apart from the mage fles contans Flckr user tags and Exchangeable Image Fle Format (Exf) nformaton. More nformaton about the exact setup of the data can be found n [4]. In the annotaton task, partcpants are asked to annotate a test set of 10,000 mages wth 99 vsual concepts. An annotated tranng set of 8,000 mages s provded. Ths mult-label learnng task [8] can be solved n three dfferent ways accordng to the type of nformaton used for learnng: 1) vsual (the mage fles), 2) textual (Flckr user tags), 3) mult-modal (vsual and textual nformaton). We developed vsual, textual and mult-modal approaches for ths task usng mult-label learnng algorthms from the Mulan open source lbrary [9]. In ths task, the relatve performance of our textual models was qute good, but that of our vsual models was bad (our group does not have expertse on computer vson), leadng to an average mult-modal (and overall) performance. In the concept-based retreval task, partcpants were gven 40 topcs consstng of logcal connectons between the 99 concepts of the photo annotaton task, such as fnd all mages that depct a small group of persons n a landscape scenery showng trees and a rver on a sunny day, along wth 2 to 5 examples mages of each topc from the tranng set of the annotaton task. Partcpants were asked to submt (up to) the 1,000 most photos for each topc n ranked order from a set of 200,000 unannotated mages. Ths task can be solved by manual constructon of the query out of the narratve of the topcs, followed by automatc retreval of mages, or by a fully automated process. We developed a manual approach that explots the mult-label models traned n the annotaton task and a fully automated query-by-example approach based on the tags of the example mages. In ths task, both our manual and automated approaches ranked 1st n all evaluaton measures by a large margn. The rest of ths paper s organzed as follows. Sectons 2 and 3 descrbe our approaches to the annotaton task and concept-based retreval task respectvely. Secton 4 presents the results of our runs for both tasks. Secton 5 concludes our work and poses future research drectons. 2 Annotaton Task Ths secton presents the vsual, textual and mult-modal approaches that we developed for the automatc photo annotaton task. There were two (eventually three) evaluaton measures to consder for ths task: a) mean nterpolated average precson (MIAP), b) example-based F-measure (F-ex), c) semantc R-precson (SR-Precson). In order to optmze a learnng approach based on each of the ntal two evaluaton measures and type of nformaton, sx models should be bult. However, there were only fve runs allowed for ths task. We therefore decded to perform model selecton based on the wdely-used mean average precson (MAP) measure for all types of nformaton. In partcular, MAP was estmated

3 usng an nternal 3-fold cross-valdaton on the 8,000 tranng mages. Our multmodal approach was submtted n three dfferent varatons to reach the total number of fve submssons. 2.1 Automatc Annotaton wth Vsual Informaton We here descrbe the approach that we followed n order to learn mult-label models usng the vsual nformaton of the mages. The flowchart of ths approach s shown n Fg. 1. x p( c x) j j Fg. 1. Automatc annotaton usng vsual nformaton. As our group does not have expertse n computer vson, we largely followed the color-descrptor extracton approach descrbed n [6,7] and used the accompanyng software tool 2 for extractng vsual features from the mages. Harrs-Laplace and Dense Samplng were used as pont detecton strateges. Furthermore seven dfferent descrptors were used: SIFT, HSV-SIFT, HueSIFT, OpponentSIFT, C-SIFT, rgsift and RGB-SIFT. For each one of the 14 combnatons of pont detecton strategy and descrptor, a dfferent codebook was created n order to obtan a fxed length representaton for all mages. Ths s also known as the bag-of-words approach. The k-means clusterng algorthm was appled to 250,000 randomly sampled ponts from the tranng set, wth the codebook sze (k) fxed to 4096 words. Fnally, we employed hard assgnment of ponts to clusters. Usng these 4,096-dmensonal vector representatons along wth the ground truth annotatons gven for the tranng mages we bult 14 mult-label tranng datasets. After expermentng wth varous mult-label learnng algorthms we found that the smple Bnary Relevance (BR) approach coupled wth Random Forests as the base classfer (number of trees = 150, number of features = 40) yelded the best results. In order to deal wth the mbalance n the number of postve and negatve examples of each label we used nstance weghtng. The weght of the examples of the mnorty class was set to (mn+maj)/mn and the weght of the examples of the majorty class was set to (mn + maj)/maj, where mn s the number of examples of the mnorty class and maj the number of examples of the majorty class. We also expermented wth sub-samplng, but the results were worse than nstance weghtng. 2 Avalable from

4 Our approach concludes wth a late fuson scheme that averages the output of the 14 dfferent mult-label models that we bult. 2.2 Automatc Annotaton wth Flckr User Tags We here descrbe the approach that we followed n order to learn mult-label models usng the tags assgned to mages by Flckr users. The flowchart of ths approach s shown n Fg. 2. Preprocessng Tags of Image Stemmer Stop Words 27,323 features Feature Selecton 4,000 features x Mult-label learnng algorthm p( c x) j j Fg. 2. Automatc annotaton usng Flckr user tags. An ntal vocabulary was constructed by takng the unon of the tag sets of all mages n the tranng set. We then appled stemmng to ths vocabulary and removed stop words. Ths led to a vocabulary of approxmately stems. The use of stemmng mproved the results, despte that some of the tags were not n the Englsh language and that we used an Englsh stemmer. We further appled feature selecton n order to remove r or redundant features and mprove effcency. In partcular, we used the χ 2 max crteron [3] to score the stems and selected the top 4000 stems, after expermentng wth a varety of szes (500, 1000, 2000, 3000, 4000, 5000, 6000 and 7000). The mult-label learnng algorthm that was found to yeld the best results n ths case was Ensemble of Classfer Chans (ECC) [5] usng Random Forests as base classfer. ECC was run wth 15 classfer chans and Random Forests wth 10 decson trees, whle all other parameters were left to ther default value. The approach that we followed to deal wth class mbalance n the case of vsual nformaton (see the prevous subsecton), was followed n ths case too. 2.3 Automatc Annotaton wth a Mult-Modal Approach Our mult-modal approach s based on a late fuson scheme that combnes the output of the 14 vsual models and the sngle textual model. The combnaton s not an average of these 15 models, because n that case the vsual models would domnate the fnal scores. Instead, we follow a herarchcal combnaton scheme. We separately average the 7 vsual models of each pont estmator and then combne the output of the textual model, the Harrs-Laplace average and the Dense Samplng average, as depcted n Fg. 3. The motvaton for ths scheme was the three dfferent vews of the mages that exsted n the data (Harrs-Laplace, Dense Samplng, user tags) as explaned n the followng two paragraphs.

5 Harrs Laplace Ensemble Model Sngle Model usng Harrs Laplace & SIFT Codebook Sngle Model usng Harrs Laplace & HSV-SIFT Codebook Sngle Model usng Harrs Laplace & HueSIFT Codebook Sngle Model usng Harrs Laplace & OpponentSIFT Codebook Averagng pc ( x) j j Sngle Model usng Harrs Laplace & C-SIFT Codebook Sngle Model usng Harrs Laplace & rgsift Codebook Sngle Model usng Harrs Laplace & RGB-SIFT Codebook Dense Samplng Ensemble Model Sngle Model usng Dense Samplng & SIFT Codebook Sngle Model usng Dense Samplng & HSV-SIFT Codebook x Image Sngle Model usng Dense Samplng & HueSIFT Codebook Sngle Model usng Dense Samplng & OpponentSIFT Codebook Averagng pc ( x) j j Averagng/ Arbtrator pc ( x) j j Sngle Model usng Dense Samplng & C-SIFT Codebook Sngle Model usng Dense Samplng& rgsift Codebook Sngle Model usng Dense Samplng& RGB-SIFT Codebook Flckr users tags Model pc ( x) j j Fg. 3. Automatc annotaton wth a mult-modal approach We can dscern two man categores of concepts n photo annotaton: objects and scenes. For objects, Harrs-Laplace performs better because t gnores the homogeneous areas, whle for scenes, Dense Samplng performs better [6]. For example, two of the concepts where Dense Samplng acheves much hgher Average Precson (AP) from Harrs-Laplace are Nght and Macro, whch are abstract, whle the nverse s happenng n concepts Fsh and Shp, whch correspond to thngs (organsms, objects) of partcular shape. Furthermore, we observe that the vsual approach performs better n concepts, such as Sky, whch for some reason (e.g. lack of user nterest for retreval by ths concept) do not get tagged. On the other hand the textual approach performs much better when t has to predct concepts, such as Horse, Insect, Dog and Baby that typcally get tagged by users. Table 1 shows the average precson for 10 concepts, half of whch sut much better the textual models and half the vsual models. Two varatons of ths scheme were developed, dfferng n how the output of the three dfferent vews s combned. The frst one, named Mult-Modal-Avg, used an averagng operator, smlarly to the one used at the lower levels of the herarchy. The second one, named Mult-Modal-MaxAP, used an arbtrator functon to select the best one out of the three outputs for each concept, accordng to nternal evaluaton results n terms of average precson. Our thrd mult-modal submsson, named Mult-Modal-MaxAP-RGBSIFT, was a prelmnary verson of Mult-Modal-MaxAP, where only the RGBSIFT descrptor was used.

6 Table 1. Average precson for 10 concepts, half of whch sut much better the textual models and half the vsual models. Concept Textual Vsual Concept Textual Vsual Arplane Trees Horse Clouds Brd Sky Insect Overexposed Dog Bg Group Thresholdng The mult-label learners used n ths work provde us wth a confdence score for each concept. Ths s fne for an evaluaton wth MIAP and SR-Precson, but does not suffce for an evaluaton wth example-based F-measure, whch requres a bpartton of the concepts nto and r ones. Ths s a typcal ssue n mult-label learnng, whch s dealt wth a thresholdng process [2]. We used the thresholdng method descrbed n [5], whch apples a common threshold across all concepts and provdes a close approxmaton of the label cardnalty (LC) of the tranng set to the predctons made on the test set. The threshold s calculated usng the followng formula: t = argmn {t 0.00,0.05,...,1.00} LC(D tran ) LC(H t (D test )) (1) where D tran s the tranng set and H t s a classfer whch has made predctons on a test set D test under threshold t. 3 Concept-Based Retreval Task We developed two dfferent approaches for the concept-based retreval task. The frst one s based on the concept relevance scores produced by the system we developed for the annotaton task. It s a manual approach, because for each topc we manually selected the topcs and manually set the strength of ther contrbuton to the fnal rankng produced by a general formula that combnes topc relevance scores. The second one s based solely on the sample mages provded for each query and s therefore fully automated. 3.1 Manual Approach Let I = 1,..., 200, 000 be the collecton of mages, Q = 1,..., 40 the set of topcs and C = 1,..., 99 the set of concepts. We frst apply our automated mage annotaton system to each mage I and obtan a correspondng 99- dmensonal vector S = [s 1, s2,..., s99 ] wth the relevance scores of ths mage to each one of the 99 concepts. For effcency reasons, we used smplfed versons of

7 our vsual approach, takng nto account only models produced wth the RGB- SIFT descrptor, whch has been found n the past to provde better results compared to other sngle color descrptors [7]. Then, based on the descrpton of each of the 40 queres, we manually select a number of concepts that we consder related to the query, ether postvely or negatvely. Formally, for topc q Q let P q C denote the set of concepts that are postvely related to q and N q C the set of concepts that are negatvely related to q, P q N q =. For each concept c n P q N q, we further defne a real valued parameter m c q 1 denotng the strength of relevance of concept c to q. The larger ths value, the stronger the nfluence of concept c to the fnal relevance score. For each topc q and mage, the scores of the concepts are combned usng (2). S q, = (s c ) m c q (1 s c ) m c q (2) c P q c N q Fnally, for each topc, we arrange the mages n descendng order accordng to the overall relevance score and we retreve a fxed number of mages (n our submssons we retreved 250 and 1,000 mages). Note that for each topc, the selecton of related concepts and the settng of values for the m c q parameters was done usng a tral-and-error approach nvolvng careful vsual examnaton of the top 10 retreved mages, as well as more relaxed vsual examnaton of the top 100 retreved mages. Two examples of topcs and correspondng combnaton of scores follow. Topc 5: rder on horse. Here we lke to fnd photos of rders on a horse. So no sculptures or pantngs are. The rder and horse can be also only n parts on the photo. It s mportant that the person s rdng a horse and not standng next to t. Based on the descrpton of ths topc and expermentaton, we concluded that concepts 75 (Horse) and 8 (Sports) are postvely related (rder on horse), whle concept 63 (Vsual Arts) s negatvely related (no sculptures or pantngs). We therefore set P 5 = {8, 75}, N 5 = {63}. All concepts were set to equal strength for ths topc: m 8,5 = m 63,5 = m 75,5 = 1. Topc 24: funny baby. We lke to fnd photos of babes lookng funny. The baby should be n the man focus of the photo and be the reason why the photo looks funny. Photos presentng funny thngs that are not related to the baby are not. Based on the descrpton of ths topc and expermentaton, we concluded that concepts 86 (Baby), 92 (Funny) and 32 (Portrat) are postvely related. We therefore set P 24 = {32, 86, 92}, N 24 =. Based on expermentaton the concept Funny was gven twce the strength of the other concepts, we set m 32,24 = m 86,24 = 1 and m 92,24 = 2. For some topcs, nstead of explctly usng the score of a group of nterrelated concepts we consdered ntroducng a vrtual concept wth score equal to the maxmum of ths group of concepts. Ths slght adaptaton of the general rule of (2), enhances ts representaton capabltes. The followng example clarfes ths adaptaton.

8 Topc 32: underexposed photos of anmals. We lke to fnd photos of anmals that are underexposed. Photos wth normal llumnaton are not. The anmal(s) should be more or less n the man focus of the mage. Based on the descrpton of ths topc and expermentaton, we concluded that concepts 44 (Anmals), 34 (Underexposed), 72 (Dog), 73 (Cat), 74 (Brd), 75 (Horse), 76 (Fsh) and 77 (Insect) are postvely related, whle concept 35 (Neutral Illumnaton) s negatvely related. The sx last specfc anmal concepts were grouped nto a vrtual concept, say concept 1001, wth score, the maxmum of the scores of these sx concepts. We then set P 32 = {34, 44, 1001}, N 32 = {35} and m 34,32 = m 44,32 = m 1001,32 = m 35,32 = 1. Fgure 4 shows the top 10 retreved mages for topcs 5, 24 and 32, along wth the Precson@10 for these topcs. 3.2 Automated Approach Apart from the narratve descrpton, each topc of the concept-based retreval task was accompaned by a set of 2 to 5 mages from the tranng set whch could be consdered for the topc. Usng these examples mages as queres we developed a Query by Example approach to fnd the most mages n the retreval set. The representaton followed the bag-of-words model and was based on the Flckr user tags assgned to each mage. To generate the feature vectors, we appled the same method as the one used for the annotaton task. Thus, each mage was represented as a 4000-dmensonal feature vector where each feature corresponds to a tag from the tranng set whch was selected by the feature selecton method. A value of 1/0 denotes the presence/absence of the tag n the tags accompanyng an mage. To measure the smlarty between the vectors representng two mages we used the Jaccard smlarty coeffcent whch s defned as the total number of attrbutes where two vectors A and B both have a value of 1 dvded by the the total number of attrbutes where ether A or B have a value of 1. Snce more than one mages where gven as examples for each topc, we added ther feature vectors n order to form a sngle query vector. Ths approach was found to work well n comparson to other approaches, such as takng only one of the example mages as query or measurng the smlarty between a retreval mage and each example mage separately and then returnng the mages from the retreval set wth the largest smlarty score to any of the queres. We attrbute ths to the fact that by addng the feature vectors, a better representaton of the topc of nterest was created whch could not be possble f only one mage (wth possbly nosy or very few tags) was consdered. As n the manual approach, we submtted two runs, one returnng the 250 and one the 1000 most smlar mages from the retreval set (n descendng smlarty order). Fgure 5 shows the top 10 retreved mages, along wth the Precson@10 for the followng topcs: Topc 10: sngle person playng a muscal nstrument. We lke to fnd pctures (no pantngs) of a person playng a muscal nstrument. The person

9 can be on stage, off stage, nsde or outsde, sttng or standng, but should be alone on the photo. It s enough f not the whole person or nstrument s shown as long as the person and the nstrument are clearly recognzable. Topc 12: snowy wnter landscaper. We lke to fnd pctures (photos or drawngs) of whte wnter landscapes wth trees. The landscape should not contan human-made objects e.g. houses, cars and persons. Only snow on the top of a mountan s not, the landscape has to be fully covered n (at least lght) snow. Topc 30: cute toys arranged to a stll-lfe. We lke to fnd photos of toys arranged to a stll-lfe. These toys should look cute n the arrangement. Smple photos of a collecton of toys e.g. n a shop are not. We see that the 10 retreved mages for topc 30 are better than those of topcs 12 and 10. Ths can be explaned by notcng that topc 12 s a dffcult one, whle the tags of the example mages for topc 10 are not very descrptve/nformatve. 4 Results We here brefly present our results, as well as our relatve performance compared to other groups and submssons. Results for all groups, as well as more detals on the data setup and evaluaton measures can be found n [4]. 4.1 Annotaton Task The offcal results of our runs are llustrated n Table 2. We notce that n terms of MIAP, the textual model s slghtly better than the vsual, whle for the other two measures, the vsual model s much better than the textual. Among the mult-modal varatons, we notce that averagng works better than arbtratng, and as expected usng all descrptors s better than usng just the RGB- SIFT one. In addton, we notce that the mult-modal approach sgnfcantly mproves over the MIAP of the vsual and textual approaches, whle t slghtly decreases/ncreases the performance of the vsual model n the two examplebased measures. Ths may partly be due to the fact that we performed model selecton based on MAP. Table 2. Offcal results of the MLKD team n the annotaton task. Run Name MIAP F-measure SR-Precson Textual Vsual Mult-Modal-Avg Mult-Modal-MaxAP-RGBSIFT Mult-Modal-MaxAP

10 Table 3 shows the rank of our best result compared to the best results of other groups and compared to all submssons. We dd qute good n terms of textual nformaton, but qute bad n terms of vsual nformaton, leadng to an overall average performance. Lack of computer vson expertse n our group may be a reason for not beng able to get results out of the vsual nformaton. Among the three evaluaton measures, we notce that overall we dd better n terms of MIAP, slghtly worse n terms of F-measure, and even worse n terms of SR-Precson. The fact that model selecton was performed based on MAP defntely played a role for ths result. Table 3. Rank of our best result compared to the best results of other teams and compared to all submssons n the annotaton task. Approach Team Rank Submsson Rank MIAP F-Measure SR-Prec MIAP F-Measure SR-Prec Vsual 9th/15 5th/15 9th/15 25th/46 12th/46 17th/46 Textual 3rd/7 2nd/7 3rd/7 3rd/8 2nd/8 4th/8 Mult-modal 5th/10 5th/10 7th/10 9th/25 7th/25 15th/25 All 5th/18 7th/18 10th/18 9th/79 19th/79 31st/ Concept-Based Retreval Task In ths task, partcpatng systems were evaluated usng the followng measures: Mean Average Precson (MAP), Precson@10, Precson@20, Precson@100 and R-Precson. The offcal results of our runs are llustrated n Table 4. We frst notce that the frst 5 runs, whch retreved 1000 mages, lead to better results n terms of MAP and R-Precson compared to the last 5 runs, whch retreved 250 mages. Obvously, n terms of Precson@10, Precson@20 and Precson@100, the results are equal. Among the manual runs, we notce that the vsual models perform qute bad. We hypothesze that a lot of concepts that favor textual rather than vsual models, as dscussed n Sect. 2, appear n most of the topcs. The textual and mult-modal models perform best, wth the Mult-Modal-Avg model havng the best result n 3 out of the 5 measures. The automated approach performs slghtly better than the vsual model of the manual approach, but stll much worse than the textual and mult-modal manual approaches. As expected, the knowledge that s provded by a human can clearly lead to better results compared to a fully automated process. However, ths s not true across all topcs, as can be seen n Table 5, whch compares the results of the best automated and manual approach for each ndvdual topc. We can see there that the automated approach performs better on 9 topcs, whle the manual on 31.

11 Table 4. Offcal results of the MLKD team n the concept-based retreval task. Run Name MAP P@10 P@20 P@100 R-Prec Manual-Vsual-RGBSIFT Automated-Textual Manual-Textual Manual-Mult-Modal-Avg-RGBSIFT Manual-Mult-Modal-MaxAP-RGBSIFT Manual-Vsual-RGBSIFT Automated-Textual Manual-Textual Manual-Mult-Modal-Avg-RGBSIFT Manual-Mult-Modal-MaxAP-RGBSIFT Table 5. Comparson of AP for each topc between automated and manual approach Topc Automated Manual Topc Automated Manual MAP Table 6 shows the rank of our best result compared to the best results of other groups and compared to all submssons. Both our manual and our automated approach ranked 1st n all evaluaton measures.

12 Table 6. Rank of our best result compared to the best results of other teams and compared to all submssons n the annotaton task. Team Rank Submsson Rank Confguratons MAP P@10 P@20 P@100 R-Prec MAP P@10 P@20 P@100 R-Prec Automated 1st/4 1st/4 1st/4 1st/4 1st/4 1st/16 1st/16 1st/16 1st/16 1st/16 Manual 1st/3 1st/3 1st/3 1st/3 1st/3 1st/15 1st/15 1st/15 1st/15 1st/15 All 1st/4 1st/4 1st/4 1st/4 1st/4 1st/31 1st/31 1st/31 1st/31 1st/31 5 Conclusons and Future Work Our partcpaton to the very nterestng photo annotaton and concept-based retreval tasks of CLEF 2011, led to a couple of nterestng conclusons. Frst of all, we found out that we need the collaboraton of a computer vson/mage processng group to acheve better results. In terms of mult-label learnng algorthms, we notced that bnary approaches worked qute well, especally when coupled wth the strong Random Forests algorthm and class mbalance ssues are taken nto account. We also reached to the concluson, that we should have performed model selecton separately for each evaluaton measure. We therefore suggest that n future versons of the annotaton task, the allowed number of submssons should be equal to the number of evaluaton measures multpled by the number of nformaton types, so that there s space n the offcal results for models wth all knds of nformaton. There s a lot of room for mprovements n the future, both n the annotaton and the very nterestng concept-based retreval task. In terms of textual nformaton, we ntend to nvestgate the translaton of non-englsh tags. We would also lke to nvestgate other herarchcal late fuson schemes, such as an addtonal averagng step for the two dfferent vsual modaltes (Harrs-Laplace, Dense Samplng) and more advanced arbtraton technques. Other thresholdng approaches for obtanng bparttons s another nterestng drecton for future study. Acknowledgments We would lke to acknowledge the student travel support from EU FP7 under grant agreement no (PetaMeda Network of Excellence). References 1. Huskes, M.J., Thomee, B., Lew, M.S.: New trends and deas n vsual concept detecton: The mr flckr retreval evaluaton ntatve. In: MIR 10: Proceedngs of the 2010 ACM Internatonal Conference on Multmeda Informaton Retreval. pp ACM, New York, NY, USA (2010)

13 2. Ioannou, M., Sakkas, G., Tsoumakas, G., Vlahavas, I.: Obtanng bparttons from score vectors for mult-label classfcaton. Tools wth Artfcal Intellgence, IEEE Internatonal Conference on 1, (2010) 3. Lews, D.D., Yang, Y., Rose, T.G., L, F.: Rcv1: A new benchmark collecton for text categorzaton research. J. Mach. Learn. Res. 5, (2004) 4. Nowak, S., Nagel, K., Lebetrau, J.: The clef 2011 photo annotaton and conceptbased retreval tasks. In: Workng Notes of CLEF 2011 (2011) 5. Read, J., Pfahrnger, B., Holmes, G., Frank, E.: Classfer chans for mult-label classfcaton. In: Proc. 20th European Conference on Machne Learnng (ECML 2009). pp (2009) 6. van de Sande, K.E.A., Gevers, T.: Unversty of Amsterdam at the Vsual Concept Detecton and Annotaton Tasks, The Informaton Retreval Seres, vol. 32: ImageCLEF, chap. 18, pp Sprnger (2010) 7. van de Sande, K.E.A., Gevers, T., Snoek, C.G.M.: Evaluatng color descrptors for object and scene recognton. IEEE Transactons on Pattern Analyss and Machne Intellgence 32(9), (2010) 8. Tsoumakas, G., Kataks, I., Vlahavas, I.: Mnng mult-label data. In: Mamon, O., Rokach, L. (eds.) Data Mnng and Knowledge Dscovery Handbook, chap. 34, pp Sprnger, 2nd edn. (2010) 9. Tsoumakas, G., Spyromtros-Xoufs, E., Vlcek, J., Vlahavas, I.: Mulan: A java lbrary for mult-label learnng. Journal of Machne Learnng Research (JMLR) 12, (July )

14 Topc 5 P@10 = 0.8 Topc 24 P@10 = 0.2 Topc 32 P@10 = 0.5 r r r r r r r r r r r r r r r Fg. 4. Retreved mages for topcs 5, 24 and 32 usng manual retreval. Images come from the MIRFLICKR-1M mage dataset [1].

15 Topc 10 = 0.3 Topc 12 = 0.2 Topc 30 = 1.0 r r r r r r r r r r r r r r r Fg. 5. Retreved mages for topcs 10, 12 and 30 usng automated retreval. Images come from the MIRFLICKR-1M mage dataset [1].

UB at GeoCLEF Department of Geography Abstract

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

More information

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

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

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

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

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

Viewpoints combined classification method in imagebased plant identification task

Viewpoints combined classification method in imagebased plant identification task Vewponts combned classfcaton method n magebased plant dentfcaton task Gábor Szűcs, Dávd Papp 2, Dánel Lovas 2 Inter-Unversty Centre for Telecommuncatons and Informatcs, Kassa str. 26., H-4028, Debrecen,

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

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

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

More information

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

Query Clustering Using a Hybrid Query Similarity Measure

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

More information

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

KIDS Lab at ImageCLEF 2012 Personal Photo Retrieval

KIDS Lab at ImageCLEF 2012 Personal Photo Retrieval KD Lab at mageclef 2012 Personal Photo Retreval Cha-We Ku, Been-Chan Chen, Guan-Bn Chen, L-J Gaou, Rong-ng Huang, and ao-en Wang Knowledge, nformaton, and Database ystem Laboratory Department of Computer

More information

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

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

More information

Performance Evaluation of Information Retrieval Systems

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

More information

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

Keywords - Wep page classification; bag of words model; topic model; hierarchical classification; Support Vector Machines

Keywords - Wep page classification; bag of words model; topic model; hierarchical classification; Support Vector Machines (IJCSIS) Internatonal Journal of Computer Scence and Informaton Securty, Herarchcal Web Page Classfcaton Based on a Topc Model and Neghborng Pages Integraton Wongkot Srura Phayung Meesad Choochart Haruechayasak

More information

Manifold-Ranking Based Keyword Propagation for Image Retrieval *

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

More information

Description of NTU Approach to NTCIR3 Multilingual Information Retrieval

Description of NTU Approach to NTCIR3 Multilingual Information Retrieval Proceedngs of the Thrd NTCIR Workshop Descrpton of NTU Approach to NTCIR3 Multlngual Informaton Retreval Wen-Cheng Ln and Hsn-Hs Chen Department of Computer Scence and Informaton Engneerng Natonal Tawan

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

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

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET 1 BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET TZU-CHENG CHUANG School of Electrcal and Computer Engneerng, Purdue Unversty, West Lafayette, Indana 47907 SAUL B. GELFAND School

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

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

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

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

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

Face Detection with Deep Learning

Face Detection with Deep Learning Face Detecton wth Deep Learnng Yu Shen Yus122@ucsd.edu A13227146 Kuan-We Chen kuc010@ucsd.edu A99045121 Yzhou Hao y3hao@ucsd.edu A98017773 Mn Hsuan Wu mhwu@ucsd.edu A92424998 Abstract The project here

More information

Improving Web Image Search using Meta Re-rankers

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

More information

Semantic Image Retrieval Using Region Based Inverted File

Semantic Image Retrieval Using Region Based Inverted File Semantc Image Retreval Usng Regon Based Inverted Fle Dengsheng Zhang, Md Monrul Islam, Guoun Lu and Jn Hou 2 Gppsland School of Informaton Technology, Monash Unversty Churchll, VIC 3842, Australa E-mal:

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

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

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

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

More information

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

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

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

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

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

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

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

Online Detection and Classification of Moving Objects Using Progressively Improving Detectors

Online Detection and Classification of Moving Objects Using Progressively Improving Detectors Onlne Detecton and Classfcaton of Movng Objects Usng Progressvely Improvng Detectors Omar Javed Saad Al Mubarak Shah Computer Vson Lab School of Computer Scence Unversty of Central Florda Orlando, FL 32816

More information

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION

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

More information

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

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

More information

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

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

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

Gender Classification using Interlaced Derivative Patterns

Gender Classification using Interlaced Derivative Patterns Gender Classfcaton usng Interlaced Dervatve Patterns Author Shobernejad, Ameneh, Gao, Yongsheng Publshed 2 Conference Ttle Proceedngs of the 2th Internatonal Conference on Pattern Recognton (ICPR 2) DOI

More information

Face Recognition using 3D Directional Corner Points

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

More information

An Evaluation of Divide-and-Combine Strategies for Image Categorization by Multi-Class Support Vector Machines

An Evaluation of Divide-and-Combine Strategies for Image Categorization by Multi-Class Support Vector Machines An Evaluaton of Dvde-and-Combne Strateges for Image Categorzaton by Mult-Class Support Vector Machnes C. Demrkesen¹ and H. Cherf¹, ² 1: Insttue of Scence and Engneerng 2: Faculté des Scences Mrande Galatasaray

More information

Backpropagation: In Search of Performance Parameters

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

More information

Pruning Training Corpus to Speedup Text Classification 1

Pruning Training Corpus to Speedup Text Classification 1 Prunng Tranng Corpus to Speedup Text Classfcaton Jhong Guan and Shugeng Zhou School of Computer Scence, Wuhan Unversty, Wuhan, 430079, Chna hguan@wtusm.edu.cn State Key Lab of Software Engneerng, Wuhan

More information

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

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

More information

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

Query classification using topic models and support vector machine

Query classification using topic models and support vector machine Query classfcaton usng topc models and support vector machne Deu-Thu Le Unversty of Trento, Italy deuthu.le@ds.untn.t Raffaella Bernard Unversty of Trento, Italy bernard@ds.untn.t Abstract Ths paper descrbes

More information

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009.

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009. Farrukh Jabeen Algorthms 51 Assgnment #2 Due Date: June 15, 29. Assgnment # 2 Chapter 3 Dscrete Fourer Transforms Implement the FFT for the DFT. Descrbed n sectons 3.1 and 3.2. Delverables: 1. Concse descrpton

More information

CSCI 5417 Information Retrieval Systems Jim Martin!

CSCI 5417 Information Retrieval Systems Jim Martin! CSCI 5417 Informaton Retreval Systems Jm Martn! Lecture 11 9/29/2011 Today 9/29 Classfcaton Naïve Bayes classfcaton Ungram LM 1 Where we are... Bascs of ad hoc retreval Indexng Term weghtng/scorng Cosne

More information

Alignment Results of SOBOM for OAEI 2010

Alignment Results of SOBOM for OAEI 2010 Algnment Results of SOBOM for OAEI 2010 Pegang Xu, Yadong Wang, Lang Cheng, Tany Zang School of Computer Scence and Technology Harbn Insttute of Technology, Harbn, Chna pegang.xu@gmal.com, ydwang@ht.edu.cn,

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

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS P.G. Demdov Yaroslavl State Unversty Anatoly Ntn, Vladmr Khryashchev, Olga Stepanova, Igor Kostern EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS Yaroslavl, 2015 Eye

More information

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

Large-scale Web Video Event Classification by use of Fisher Vectors

Large-scale Web Video Event Classification by use of Fisher Vectors Large-scale Web Vdeo Event Classfcaton by use of Fsher Vectors Chen Sun and Ram Nevata Unversty of Southern Calforna, Insttute for Robotcs and Intellgent Systems Los Angeles, CA 90089, USA {chensun nevata}@usc.org

More information

Relevance Feedback for Image Retrieval

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

More information

Multi-modal Multi-label Semantic Indexing of Images Based on Hybrid Ensemble Learning

Multi-modal Multi-label Semantic Indexing of Images Based on Hybrid Ensemble Learning Mult-modal Mult-label Semantc Indexng of Images Based on Hybrd Ensemble Learnng We L 1, Maosong Sun 1, and Chrstopher Habel 2 1 State Key Lab of Intellgent Technology and Systems Department of Computer

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

Decision Strategies for Rating Objects in Knowledge-Shared Research Networks

Decision Strategies for Rating Objects in Knowledge-Shared Research Networks Decson Strateges for Ratng Objects n Knowledge-Shared Research etwors ALEXADRA GRACHAROVA *, HAS-JOACHM ER **, HASSA OUR ELD ** OM SUUROE ***, HARR ARAKSE *** * nsttute of Control and System Research,

More information

Enhancement of Infrequent Purchased Product Recommendation Using Data Mining Techniques

Enhancement of Infrequent Purchased Product Recommendation Using Data Mining Techniques Enhancement of Infrequent Purchased Product Recommendaton Usng Data Mnng Technques Noraswalza Abdullah, Yue Xu, Shlomo Geva, and Mark Loo Dscplne of Computer Scence Faculty of Scence and Technology Queensland

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

Fuzzy Modeling of the Complexity vs. Accuracy Trade-off in a Sequential Two-Stage Multi-Classifier System

Fuzzy Modeling of the Complexity vs. Accuracy Trade-off in a Sequential Two-Stage Multi-Classifier System Fuzzy Modelng of the Complexty vs. Accuracy Trade-off n a Sequental Two-Stage Mult-Classfer System MARK LAST 1 Department of Informaton Systems Engneerng Ben-Guron Unversty of the Negev Beer-Sheva 84105

More information

Deep Classification in Large-scale Text Hierarchies

Deep Classification in Large-scale Text Hierarchies Deep Classfcaton n Large-scale Text Herarches Gu-Rong Xue Dkan Xng Qang Yang 2 Yong Yu Dept. of Computer Scence and Engneerng Shangha Jao-Tong Unversty {grxue, dkxng, yyu}@apex.sjtu.edu.cn 2 Hong Kong

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

Random Kernel Perceptron on ATTiny2313 Microcontroller

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

More information

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

Analysis of Continuous Beams in General

Analysis of Continuous Beams in General Analyss of Contnuous Beams n General Contnuous beams consdered here are prsmatc, rgdly connected to each beam segment and supported at varous ponts along the beam. onts are selected at ponts of support,

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 enhanced representation of time series which allows fast and accurate classification, clustering and relevance feedback

An enhanced representation of time series which allows fast and accurate classification, clustering and relevance feedback An enhanced representaton of tme seres whch allows fast and accurate classfcaton, clusterng and relevance feedback Eamonn J. Keogh and Mchael J. Pazzan Department of Informaton and Computer Scence Unversty

More information

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

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

More information

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

Feature Selection as an Improving Step for Decision Tree Construction

Feature Selection as an Improving Step for Decision Tree Construction 2009 Internatonal Conference on Machne Learnng and Computng IPCSIT vol.3 (2011) (2011) IACSIT Press, Sngapore Feature Selecton as an Improvng Step for Decson Tree Constructon Mahd Esmael 1, Fazekas Gabor

More information

Multiclass Object Recognition based on Texture Linear Genetic Programming

Multiclass Object Recognition based on Texture Linear Genetic Programming Multclass Object Recognton based on Texture Lnear Genetc Programmng Gustavo Olague 1, Eva Romero 1 Leonardo Trujllo 1, and Br Bhanu 2 1 CICESE, Km. 107 carretera Tjuana-Ensenada, Mexco, olague@ccese.mx,

More information

SRBIR: Semantic Region Based Image Retrieval by Extracting the Dominant Region and Semantic Learning

SRBIR: Semantic Region Based Image Retrieval by Extracting the Dominant Region and Semantic Learning Journal of Computer Scence 7 (3): 400-408, 2011 ISSN 1549-3636 2011 Scence Publcatons SRBIR: Semantc Regon Based Image Retreval by Extractng the Domnant Regon and Semantc Learnng 1 I. Felc Raam and 2 S.

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Decson surface s a hyperplane (lne n 2D) n feature space (smlar to the Perceptron) Arguably, the most mportant recent dscovery n machne learnng In a nutshell: map the data to a predetermned

More information

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

Classifying Acoustic Transient Signals Using Artificial Intelligence

Classifying Acoustic Transient Signals Using Artificial Intelligence Classfyng Acoustc Transent Sgnals Usng Artfcal Intellgence Steve Sutton, Unversty of North Carolna At Wlmngton (suttons@charter.net) Greg Huff, Unversty of North Carolna At Wlmngton (jgh7476@uncwl.edu)

More information

Network Intrusion Detection Based on PSO-SVM

Network Intrusion Detection Based on PSO-SVM TELKOMNIKA Indonesan Journal of Electrcal Engneerng Vol.1, No., February 014, pp. 150 ~ 1508 DOI: http://dx.do.org/10.11591/telkomnka.v1.386 150 Network Intruson Detecton Based on PSO-SVM Changsheng Xang*

More information

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010 Smulaton: Solvng Dynamc Models ABE 5646 Week Chapter 2, Sprng 200 Week Descrpton Readng Materal Mar 5- Mar 9 Evaluatng [Crop] Models Comparng a model wth data - Graphcal, errors - Measures of agreement

More information

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

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

More information

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

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

More information

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

Arabic Text Classification Using N-Gram Frequency Statistics A Comparative Study

Arabic Text Classification Using N-Gram Frequency Statistics A Comparative Study Arabc Text Classfcaton Usng N-Gram Frequency Statstcs A Comparatve Study Lala Khresat Dept. of Computer Scence, Math and Physcs Farlegh Dcknson Unversty 285 Madson Ave, Madson NJ 07940 Khresat@fdu.edu

More information

A Feature-Weighted Instance-Based Learner for Deep Web Search Interface Identification

A Feature-Weighted Instance-Based Learner for Deep Web Search Interface Identification Research Journal of Appled Scences, Engneerng and Technology 5(4): 1278-1283, 2013 ISSN: 2040-7459; e-issn: 2040-7467 Maxwell Scentfc Organzaton, 2013 Submtted: June 28, 2012 Accepted: August 08, 2012

More information

Machine Learning. Topic 6: Clustering

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

More information

Fingerprint matching based on weighting method and SVM

Fingerprint matching based on weighting method and SVM Fngerprnt matchng based on weghtng method and SVM Ja Ja, Lanhong Ca, Pnyan Lu, Xuhu Lu Key Laboratory of Pervasve Computng (Tsnghua Unversty), Mnstry of Educaton Bejng 100084, P.R.Chna {jaja}@mals.tsnghua.edu.cn

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

CHAPTER 2 DECOMPOSITION OF GRAPHS

CHAPTER 2 DECOMPOSITION OF GRAPHS CHAPTER DECOMPOSITION OF GRAPHS. INTRODUCTION A graph H s called a Supersubdvson of a graph G f H s obtaned from G by replacng every edge uv of G by a bpartte graph,m (m may vary for each edge by dentfyng

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

Heterogeneous Visual Features Fusion via Sparse Multimodal Machine

Heterogeneous Visual Features Fusion via Sparse Multimodal Machine 013 IEEE Conference on Computer Vson and Pattern Recognton Heterogeneous Vsual Features Fuson va Sparse Multmodal Machne Hua ang, Fepng Ne, Heng Huang, Chrs Dng Department of Electrcal Engneerng and Computer

More information

Related-Mode Attacks on CTR Encryption Mode

Related-Mode Attacks on CTR Encryption Mode Internatonal Journal of Network Securty, Vol.4, No.3, PP.282 287, May 2007 282 Related-Mode Attacks on CTR Encrypton Mode Dayn Wang, Dongda Ln, and Wenlng Wu (Correspondng author: Dayn Wang) Key Laboratory

More information

PYTHON IMPLEMENTATION OF VISUAL SECRET SHARING SCHEMES

PYTHON IMPLEMENTATION OF VISUAL SECRET SHARING SCHEMES PYTHON IMPLEMENTATION OF VISUAL SECRET SHARING SCHEMES Ruxandra Olmd Faculty of Mathematcs and Computer Scence, Unversty of Bucharest Emal: ruxandra.olmd@fm.unbuc.ro Abstract Vsual secret sharng schemes

More information

A Novel Term_Class Relevance Measure for Text Categorization

A Novel Term_Class Relevance Measure for Text Categorization A Novel Term_Class Relevance Measure for Text Categorzaton D S Guru, Mahamad Suhl Department of Studes n Computer Scence, Unversty of Mysore, Mysore, Inda Abstract: In ths paper, we ntroduce a new measure

More information

Video Classification and Retrieval with the Informedia Digital Video Library System

Video Classification and Retrieval with the Informedia Digital Video Library System Vdeo Classfcaton and Retreval wth the Informeda Dgtal Vdeo Lbrary System A. Hauptmann, R. Yan, Y. Q, R. Jn, M. Chrstel, M. Derthck, M.-Y. Chen, R. Baron, W.-H. Ln, and T. D. Ng. Carnege Mellon Unversty,

More information