A Bayesian Framework for Fusing Multiple Word Knowledge Models in Videotext Recognition

Size: px
Start display at page:

Download "A Bayesian Framework for Fusing Multiple Word Knowledge Models in Videotext Recognition"

Transcription

1 A Bayesan Framework for Fusng Multple Word Knowledge Models n Vdeotext Recognton DongQng Zhang and Shh-Fu Chang Department of Electrcal Engneerng, Columba Unversty New York, NY 0027, USA. {dqzhang, sfchang}@ee.columba.edu Abstract Vdeotext recognton s challengng due to low resoluton, dverse fonts/styles, and cluttered background. Past methods enhanced recognton by usng multple frame averagng, mage nterpolaton and lexcon correcton, but recognton usng mult-modalty language models has not been explored. In ths paper, we present a formal Bayesan framework for vdeotext recognton by combnng multple knowledge usng mxture models, and descrbe a learnng approach based on Expectaton-Maxmzaton (EM). In order to handle unseen words, a back-off smoothng approach derved from the Bayesan model s also presented. We exploted a prototype that fuses the model from closed capton and that from the Brtsh Natonal Corpus. The model from closed capton s based on a unque tme dstance dstrbuton model of vdeotext words and closed capton words. Our method acheves a sgnfcant performance gan, wth word recognton rate of 76.8% and character recognton rate of 86.7%. The proposed methods also reduce false vdeotext detecton sgnfcantly, wth a false alarm rate of 8.2% wthout substantal loss of recall. Keywords: Vdeotext recognton, Vdeo OCR, Vdeo ndexng, Informaton Fusng. Multmodal Recognton.. Introducton Vdeotext recognton s dffcult due to low resoluton, dverse fonts, sze, colors, styles, and cluttered background. There are two categores of vdeotext n dgtal vdeos: overlay text, whch s added by vdeo edtors; scene text, whch s embedded n real-world obects. Although overlay text and scene text share some common propertes, overlay text s easer to detect than scene text n general and s the focus of ths paper. A complete vdeotext recognton system nvolves both ssues of detecton and recognton. Vdeotext detecton has been extensvely studed n recent years [,2,3,4], but vdeotext recognton s much less explored. Some relevant works n vdeotext recognton nclude template matchng [], SVM classfer [5], and those usng document OCR engnes [2] etc. Enhancement schemes have been studed by many researchers, for example, temporal averagng of multple frame [,4], spatal nterpolaton [4], font context [6] and word correcton by dctonary []. But the potental of usng language models, especally multmodal models, has not been explored. The most related dea s word correcton usng edt dstance by dctonary []. But such method works well only when the character recognton error rate s low. The language model has been wdely adopted n speech recognton [7] and handwrtten recognton [8]. To construct a language model, one needs text corpora contanng a large number of text documents. The problem encountered by vdeotext recognton s the dffculty n acqurng suffcent data from vdeos for language model constructon. Language models can be created from general lngustc corpora, but t may be naccurate. Recognton usng multmodalty s another way to enhance performance. Today s broadcast vdeos usually are assocated wth many text sources, such as closed captons, and onlne web documents. These documents can be used to enhance the vdeotext recognton, snce they contan words whch are often related to words n vdeotext. However, solely relyng on external document source s not suffcent. Take the example of closed capton, only about 40% to 50% of vdeotext words can be found n closed capton. Therefore, there s great promse n combnng language models from dfferent sources wth dfferent modaltes. Ths paper ams at ths problem by constructng a Bayesan framework to fuse the word knowledge models from multple sources. The framework s establshed usng mxture models and ts tranng approach s derved from the Expectaton-Maxmzaton (EM) algorthm. In order to ncrease the recognton performance of characters and unseen words, a smoothng scheme s derved to back-off the word recognton to the baselne character recognton approach. To valdate the framework n the practcal doman, we use the closed captons n vdeos and lngustc corpus to extract the multple word knowledge models. The knowledge model from closed capton s bult by learnng a unque dstrbuton model of the tme dstance between the vdeotext and ther matched counterpart n closed capton.

2 The general lngustc knowledge model s extracted from the Brtsh Natonal Corpus. We also developed a multple frame ntegraton technque as a post processng stage. Besdes usng multple frame averagng [], we explored a multple frame votng scheme, whch frst dentfy dentcal text blocks n dfferent frames, then use votng process to select the domnant word recognton output among the text blocks. Fgure shows our system dagram for vdeotext recognton fusng multple word knowledge models. We evaluate the system on sx news vdeos from three dfferent channels wth about 200 vdeotext words. The experments showed a 5% accuracy mprovement comparng the proposed method wth the baselne technque. The combned model also performs better than ndvdual models by 4.4%. When used as a postprocessng step, the word recognton technque plus temporal votng also help reduce vdeotext detecton false alarms sgnfcantly. The paper s organzed as follows: Secton 2 brefly descrbes the pre-processng approaches ncludng detecton, bnarzaton and segmentaton. Secton 3 presents the baselne character recognton system. Secton 4 descrbes the Bayesan framework for word recognton. Secton 5 presents a prototype model usng closed capton and the Brtsh Natonal Corpus. Secton 6 descrbes experments wth the results. Text Detect & Vdeo Segmentaton Input Refne Segmentaton Baysan Word Knowledge Models Character Lkelhood >>WB News.. > Today s the thrd Speech closed Transcrpt capton Bayesan Word Recognton Lngustc Corpus Vdeotext Words Output Fgure. Flowchart of the proposed vdeotext recognton system. It fuses word knowledge models from closed capton and lngustc corpus. 2. Pre-processng We frst brefly descrbe the preprocessng stage ncludng vdeotext detecton, bnarzaton and segmentaton. Careful desgn of these modules s mportant for later robust recognton processes. 2.. Vdeotext detecton We use the vdeotext detecton algorthm developed n our pror works [9,0] to extract the vdeotexts from the vdeos. The system frst computes texture and moton features by usng the DCT coeffcents and moton vectors drectly from the MPEG compressed streams wthout full decodng. These features are used to detect canddate vdeotext regons, wthn each of whch color layerng s used to generate several hypothetcal bnary mages. Afterwards a groupng process s performed n each bnary mage to fnd the character blocks. Fnally, a layout analyss process s appled to verfy the layout of these character blocks usng a rule-based approach Bnarzaton and character segmentaton Bnarzaton and character segmentaton s dffcult for vdeotext due to color varaton and small spacng between characters. We developed teratve and multhypothess methods to handle these problems. Fxed threshold value s not suted for vdeotext bnarzaton, because vdeotext ntensty may show great varatons. We developed an teratve threshold selecton method to dynamcally adust the ntensty threshold value untl the broken strokes of characters appear. Such dea s smlar to that proposed n []. The character segmentaton method s based on local mnma searchng n the vertcal proecton profle [0]. A segmentaton lne s dentfed by thresholdng the proecton profle. To reduce the recognton errors caused by character segmentaton, multple segmentaton hypotheses are used to produce canddate characters. Pror work n [] searched for the optmal hypothess usng dynamc programmng. In our case, snce the number of canddate segmentaton ponts s usually small (one to twenty, mostly less than four), an exhaustve search s performed. Word segmentaton s needed to fnd complete word segments for recognton. To realze ths, the medan value of the character spacng s frst calculated. If the spacng between two characters s larger than two tmes the medan value, the segmentaton lne s marked as a word boundary. 3. Character recognton The character recognton step nvolves the feature extracton from a sngle character and shapng of character condtonal densty functons. 3.. Character feature extracton The feature set for character recognton nclude {Zernke magntude; Drecton proporton; st -order perpheral features; second-order perpheral features; vertcal proecton profle; horzontal proecton profle.} These features are selected from a larger feature set manually. For Zernke moment features, readers are referred to the paper [2] for complete descrpton. And the descrpton of other features can be found n [3]. These features lead to an overall dmenson of Character condtonal densty functon The character condton densty functon s modeled usng Parzen wndow [4]. One can also use Gaussan Mxture Model (GMM). However, the GMM has the

3 overfttng problem when the dmenson of data space s hgh. Regularzaton can be ntroduced to handle ths problem, such as usng Bayesan penalty term [4]. However, n our experments, we found that the Parzen wndow approach outperforms the regualzed GMM. For Parzen Wndow, the sample ponts are generated usng a dstorton modelng procedure. We apply a varety of geometrcal dstortons to each standard font mage to obtan tranng samples. Dstortons nclude 4 fonts, 3 aspect ratos, 3 character weghts, and 5 szes. The sze varaton has lttle mpact on recognton, therefore we average the feature vectors correspondng to dfferent szes. Ths leads to 36 sample data for each character. A Gaussan kernel s used for the Parzen wndow method. The densty functon can be adusted by changng the varance of the Gaussan kernel. In order to maxmze the character recognton performance, the varance of the Gaussan kernel needs to be tuned usng tranng data. Gven a feature vector for a character mage, a baselne system for the character recognton s to compare the lkelhood values of dfferent characters and select the character correspondng to the hghest lkelhood. We wll refer to ths method as baselne character recognton method throughout the paper. 4. Bayes word recognton framework The vdeotext word recognton problem can be formulated usng Bayesan method or the maxmum a- posteror (MAP) recognton as: wˆ = arg max w x) w () = arg max log x w) + log w) w [ ] where x s the word feature vector, and w s a canddate word. x w) s called word observaton model constructed from the character condtonal densty functon. p (w) s called language model n the communty of speech recognton. It specfes the pror probably of each word. Here we not only use the lngustc corpus but also the models from other sources, such as closed capton, thus we call w) as Word Knowledge Model (WKM). 4.. Word condtonal densty functon Word observaton model s constructed from the sngle character condtonal densty functon. Suppose after segmentaton, N character mages are segmented from a word mage, and the feature vectors of these characters are x, x2,..., xn. Then the constructed word feature vector s x = { x, x2,..., x N }. The word observaton model denoted by word lkelhood functon therefore s: N x w) = x w) = x c ), = = 0, N = w = N w N where c Α, Α s the alphabet, whch currently nclude 62 characters (26 letters wth lower and upper case, plus 0 dgts) Fusng multple word knowledge models As dscussed earler, the language model w) could be obtaned by usng lngustc corpus; but t may be naccurate due to the lmt of tranng data. Combnaton of multple models could be a remedy to ths problem by addng other relevant knowledge nto the general model. These addtonal sources can be acqured easly n today s dstrbuted nformaton envronment, for example closed capton n the vdeo stream, or onlne documents on related web stes etc. Suppose that we have obtaned or learned the language models from dfferent sources. We denote such models as p ( w K ), where K denotes the nformaton source. Each word knowledge model (WKM) covers a subset of all possble words. Suppose the subset covered by each model s S. We use a lnear combnaton of these WKMs to form a mxture word knowledge model: subect to: = C = K (2) w) α w ) (3) p ( w K ) = and α = (4) w S C = where C s the number of sources. The combned model wll cover all the words that belong to S = U S. To obtan the weghng vector a = { α α,..., }, we, 2 use the vdeotexts n the tranng set. The data needed for tranng s much less than that requred by constructng a general language model due to the small sze of the parameter space. The optmal weghtng vector should maxmze the ont probablty of the tranng set based on the maxmum lkelhood tranng,.e.: = arg max p ( w α) a, subect to α = (5) α w T c = where T s the tranng set. Although ths s a standard constrant maxmzaton problem, t s actually dffcult to solve and get the closed form soluton. However, t can be solved teratvely by usng the Expectaton-Maxmzaton (EM) method [5]. The update equatons based on EM s: old w K ) α N w ) = C, α = w ) (6) old N = w K ) α = α c

4 where N s the number of the tranng vdeotext words Recognton of unseen word The combned model s usually unable to cover all vdeotext words n a vdeo. For nstance, n news vdeo, about 5% vdeotext words cannot be found n ether closed capton or lngustc corpus. Drectly applyng language models to those unseen words may change correctly recognzed words and thus ncrease recognton errors. To handle ths problem, we use a method to back-off the word recognton process to character recognton n certan condton. Such method has been used n speech recognton [6]. We derve the back-off method based on the Bayesan recognton framework, where the word knowledge model s decomposed as: w) = w w S) w S) + w w S) w S) = w w S)( c) + w w S) c where S denotes the word dctonary covered by the knowledge models, and c s the probablty that a vdeotext word falls out of S. w w S) s the pror specfed by the knowledge model. For an unseen word, ths term wll be zero. w w S) s a hypothetcal dstrbuton for all words that are not n S, whose value s zero for all seen words. Based on these, we wll get the followng back-off condton when consderng a canddate word w S from S: [ log w ) + log w w S) ] log x w ) d S S S < S (7) x (8) where w s the baselne character recognton output. d S s the back-off threshold, whch can be traned usng a straghtforward method. Note that back-off s only appled when the baselne recognton result cannot be found n S. The dervaton of ths equaton and estmaton of d usng the tranng set s dscussed n [7] Post-processng A unque nature of vdeotext comparng wth document mage text s the redundancy of text mages: capton text usually stays on the vdeo for a few seconds, resultng n duplcates of the same text blocks n consecutve frames. Pror systems employed temporal averagng to take advantage of such redundancy. However, we found that although temporal averagng s able to reduce the addtve nose, t cannot fully avod the false recognton caused by segmentaton error or character corrupton by background perturbaton. Thus we propose to use a multple frame votng method usng recognton results from each ndvdual frame. To realze ths, we group smlar text blocks wthn a local temporal wndow together. The smlarty s measured usng sum of pxel-to-pxel color dstances between each vdeotext blocks. The votng process s performed by selectng the most frequent output among all word recognton output n the same group. Such algorthm effectvely elmnates the false recognton due to erroneous character segmentaton. The above temporal votng process not only mproves the word recognton accuracy, but also mproves the text detecton accuracy. Detecton false alarm s fltered out f the posteror of a word s lower than certan threshold before votng, or the word count of the most frequent word n a group s one. 5. Prototype usng closed capton and Brtsh natonal corpus We realze a prototype of the proposed framework and algorthms by usng the closed capton (CC) stream assocated wth the vdeo and an external lngustc corpus, Brtsh Natonal Corpus (BNC). The multple knowledge models under these two sources can be wrtten as: w) = α w CC) + α w BNC) (9) cc 5.. Buldng the word knowledge model from closed capton For a word drawn from the CC model, ts pror s assumed to only depend on two factors: () the tme dstance between the CC word and the vdeotext (VT) word beng recognzed, t = t sw t vw, and (2) The part-ofspeech (POS) of the CC word, S. Words far from the vdeotext word have lower pror probabltes. Words of dfferent POS (e.g., verb vs. noun) have dfferent prors of appearng n the vdeotext. Usng CC we can construct a CC wordlst (CCW), whch ncludes all words that occur n CC at least once. If there are multple nstances of a word, CCW only keeps one nstance. Thus we model the followng word pror: p ( w = w' CC) = w = w t, S) (0) C w = w' where w s the word n CCW that may appear multple tmes, w s the -th nstance of the word w appearng n the CC stream, BNC t, S s the tme dstance and POS of w respectvely, and C s a normalzaton constant C = w' CCW w = w' w = w t, S ) () Because tranng such model s complcated due to the presence of POS, we use a smplfed model: when the POS of w s a stop word or preposton, the probablty s zero, otherwse t depends solely on the tme dstance. In other words, only non-stop word and non-preposton words are consdered n tranng and recognton. The lkelhood functon can be n varous functon forms, whch can be determned by comparng the emprcal dstrbuton and the estmated dstrbuton usng

5 Ch-square hypothess test. We used two hypotheses: Gaussan functon and exponental functon. The hypothess test shows that the exponental functon s closer to the emprcal dstrbuton. The non-causal exponental tme dstance densty functon we adopted s as follows: t λ w = w t, S SP) = e (2) 2λ SP denotes stop word or proposton word. Ths s a double exponental model (DPM). For a causal model, t s straghtforward to modfy the equaton (2) by removng the rght tal of the DPM. To tran the parameter of ths model, a standard maxmum lkelhood approach s used [5] usng the pool of all matched word pars n CC and vdeotext. Based on our experments, λ l = provdes satsfactory results. Accordng to CC model, a word that cannot be found n CC wll be assgned a zero probablty of p ( w CC) Knowledge model from BNC Word knowledge model s also extracted from the Brtsh Natonal Corpus (BNC) [8]. Brtsh Natonal Corpus ncludes a large number of text documents for tranng language models. BNC also provdes the word lsts wth the use frequency of each word. We use the wrtten Englsh verson lsts contanng about 200,000 words. The lst ncludes all nflected forms of each word stem as well as ther frequency. In vdeotext, stop words are rarely used, but they hold hghest frequency n the BNC word lst. In order to get a more accurate dstrbuton functon, the word frequences of these stop words are manually re-assgned to a small value. After these processes, the word knowledge model extracted from the BNC s the normalzed verson of the word frequency: P ( w BNC) = Freq( w) Freq( w ) (3) w BNC There s spellng dfference between Brtsh words and Englsh words [9]. However, n BNC word lst, both Brtsh spellng and Amercan spellng are ncluded [8] for most words. In experments, we confrmed spellng dfferences dd not result n performance degradaton. 6. Experments Our experment data nclude sx news vdeos from three channels broadcasted on dfferent days. The vdeos nclude dfferent stores, dfferent fonts and ntensty values of vdeotext. The format of the vdeos s MPEG- wth SIF resoluton (352x240 pxels). The overall duraton of the test set s about sx hours. A cross valdaton process s used to evaluate the algorthms. That s, the methods are traned usng vdeos from two channels and are tested usng the vdeos from remanng channels. Durng the tranng process, the estmated parameter set ncludes parameters of the tme dstance dstrbuton for the closed capton model, the weghtng vectors of the mxture model, and the back-off threshold. The varance of the Gaussan kernel for the Parzen wndow s also determned emprcally usng the tranng set. In the testng stage, the detecton program s frst carred out to detect the super-mposed text blocks. The overall detecton recall rate s 97% and the ntal precson rate of detecton s 70%. The detected text blocks are then passed to bnarzaton, segmentaton, recognton, and post-processng. After word recognton and post-processng, the false detectons are fltered, leadng to an mproved precson rate of 9.8% wth degraded recall rate of 95.6 %. The performance of recognton wthn the correctly detected set s shown n Table. Here one word recognton error s counted as long as there are one or more character recognton errors n the word. The mprovement n character recognton s large (+9.6%); the mprovement n word accuracy s even more sgnfcant (+5%). Table. Recognton Accuracy Vdeos Char Accuracy Word Accuracy B K B K w#: % 86.7% 25.8% 76.8% Legends: B: baselne character recognton, K: Knowledge based recognton, w#: total number of words. Fgure 2 shows some examples of the vdeotext recognton results, wth dfferent types of success and falure grouped together. Under each text mage, two recognton results are shown the left one shows the result usng the baselne method whle the rght one shows the result usng the knowledge-based recognton method combnng both BNC and CC models. The one n the bold face s the fnal result selected by our system usng the back-off procedure descrbed n Secton 4.3. Table 2. Contrbuton of CC and BNC Vdeos BNC CC CC+BNC CC Cont w#: % 48.6% 76.8% 4.4% Legends: BNC: use BNC only, CC: use CC only, CC+BNC: use both BNC and CC, CC Cont: CC We also conducted separate tests to study the ndvdual contrbutons from each knowledge model. In table 2, the BNC column shows the performance usng the BNC model only, the CC column shows the performance usng the closed capton model only, the CC+BNC column shows the results combnng both models. The results show that when used alone, the BNC model s

6 untop8bttabtb unforgettable celbforvb calforna ph34delph ahstorcal paqmbnt payment; benl8e qlovep doachlm dense glover oachm; (a) ttm6 tme; pr0tests Pl8nnbd protests planned (b) grape scandal grade scandal schuylerylle secretaryshp a3sst8nt pdq8ad Assstant adabas; Yell us at ten mlneola mdsole dclwntown brookt chnatown brooks tell us at ten (c) (d) (e) Fgure 2. (a) Some results of knowledge models (b) Recognton of vdeotexts wth varous styles (c) False recognton corrected by the surroundng CC words (d) Back-off trggered due to unseen words (e) False recognton due to poor segmentaton and thresholdng. more effectve than the CC model. When they are combned, the CC model adds 4.4% accuracy mprovement on top of the result usng the BNC model only. When we further analyzed the data, we found the gan most came from the refnement to the word pror probablty. Fgure 2(c) shows several examples of errors corrected by addng the CC model. 7. Concluson We have developed a Bayesan framework for vdeotext recognton, n whch the pror probabltes of words are estmated by combnng multple word knowledge models. Our current prototype ncludes synchronzed closed capton and lngustc corpus, Brtsh Natonal Corpus, as knowledge models. We used an EM based method for learnng the fusng model. We have also developed a back-off process to handle unseen words n the model. To estmate the prors for words n the closed captons, we used an effectve statstcal model takng nto account the tme dstances of the closed capton word to the vdeotext. The experments show that such multmodalty knowledge fusng method results n sgnfcant performance gan. When combnng the word recognton and temporal votng n a post-processng stage, the false detecton of text detecton s also sgnfcantly reduced. 8. References [] T. Sato, T. Kanade, E. Hughes, and M. Smth, "Vdeo OCR: Indexng Dgtal News Lbrares by Recognton of Supermposed Captons", Multmeda Systems, 7: , 999. [2] R. Lenhart, W. Effelsberg, "Automatc text segmentaton and text recognton for vdeo ndexng", Multmeda System, [3] J.C. Shm, C. Dora and R. Bolle, Automatc Text Extracton from Vdeo for Content-Based Annotaton and Retreval", Proc. 4th Internatonal Conference on Pattern Recognton, volume, pp , Brsbane, Australa, August 998. [4] H. L; D. Doermann, O. Ka, Automatc text detecton and trackng n dgtal vdeo, IEEE Trans. on Image Processng, Vol 9, No., January [5] C. Dora, H. Aradhye, J.C. Shm, End-to-End Vdeotext Recognton for Multmeda Content Analyss. IEEE Conference on Multmeda and Exhbton (ICME 200). [6] H. Aradhye, C. Dora, J.C. Shm, Study of Embedded Font Context and Kernel Space Methods for Improved Vdeotext Recognton, IEEE Internatonal Conference on Image Processng (ICIP 200). [7] B. Gold, N. Morgan, Speech and Audo processng, John Wley & Sons, Inc (999). [8] R. Plamondon, S.N. Srhar, On-Lne and Off-Lne Handwrtng Recognton: A comprehensve Survey, IEEE Trans. on PAMI, Vol. 22, No., Janury [9] D. Zhang, and S.F. Chang, "Accurate Overlay Text Extracton for Dgtal Vdeo Analyss", Columba Unversty Advent Group Techncal Report 2003 #005. [0] D. Zhang, R.K. Raendran and S.F. Chang, "General and Doman-specfc Technques for Detectng and Recognzng Supermposed Text n Vdeo", Proceedng of Internatonal Conference on Image Processng, Rochester, New York, USA. [] T. Rdler, S. Calvard, "Pcture Thresholdng Usng an Iteratve Selecton Method", IEEE transactons on Systems, Man and Cybernetcs, August, 978. [2] A. Khotanzad and Y.H. Hong, Invarant Image Recognton by Zernke Moments, IEEE Transactons on Pattern Analyss and Machne Intellgence, Vol 2, No 5, May 990. [3] R. Romero, D. Touretzkey, and R.H. Thbadeau, Optcal Chnse Character Recognton Usng Probablstc Neural Networks, CMU Techncal Report. [4] D. Ormonet and V. Tresp. Improved Gaussan Mxture Densty Estmates Usng Bayesan Penalty Terms and Network Averagng. In Advances n Neural Informaton Processng Systems, volume 8, The MIT Press, 996. [5] R.O. Duda, P.E. Hart, D.G. Stock, Pattern Classfcaton. Wley-Interscence, New York, NY, 2 ed., [6] S.M. Katz. Estmaton of Probabltes from Sparse Data for the Language Model Component of a Speech Recognzer. IEEE Trans. on Acoustcs, Speech and Sgnal Processng, 35(3): , 987. [7] D. Zhang, S.F. Chang, A Mult-Model Bayesan Framework for Vdeotext Recognton, ADVENT Techncal Report 2003, Columba Unversty. [8] Brtsh Natonal Corpus, Web homepage: [9] Dctonary of Amercan and Brtsh Us(e)age. URL:

Accurate Overlay Text Extraction for Digital Video Analysis

Accurate Overlay Text Extraction for Digital Video Analysis Accurate Overlay Text Extracton for Dgtal Vdeo Analyss Dongqng Zhang, and Shh-Fu Chang Electrcal Engneerng Department, Columba Unversty, New York, NY 10027. (Emal: dqzhang, sfchang@ee.columba.edu) Abstract

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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 Background Subtraction for a Vision-based User Interface *

A Background Subtraction for a Vision-based User Interface * A Background Subtracton for a Vson-based User Interface * Dongpyo Hong and Woontack Woo KJIST U-VR Lab. {dhon wwoo}@kjst.ac.kr Abstract In ths paper, we propose a robust and effcent background subtracton

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

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

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

Learning a Class-Specific Dictionary for Facial Expression Recognition

Learning a Class-Specific Dictionary for Facial Expression Recognition BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 16, No 4 Sofa 016 Prnt ISSN: 1311-970; Onlne ISSN: 1314-4081 DOI: 10.1515/cat-016-0067 Learnng a Class-Specfc Dctonary for

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

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

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

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

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

Unsupervised Learning

Unsupervised Learning Pattern Recognton Lecture 8 Outlne Introducton Unsupervsed Learnng Parametrc VS Non-Parametrc Approach Mxture of Denstes Maxmum-Lkelhood Estmates Clusterng Prof. Danel Yeung School of Computer Scence and

More information

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

Recognition of Handwritten Numerals Using a Combined Classifier with Hybrid Features

Recognition of Handwritten Numerals Using a Combined Classifier with Hybrid Features Recognton of Handwrtten Numerals Usng a Combned Classfer wth Hybrd Features Kyoung Mn Km 1,4, Joong Jo Park 2, Young G Song 3, In Cheol Km 1, and Chng Y. Suen 1 1 Centre for Pattern Recognton and Machne

More information

Incremental MQDF Learning for Writer Adaptive Handwriting Recognition 1

Incremental MQDF Learning for Writer Adaptive Handwriting Recognition 1 200 2th Internatonal Conference on Fronters n Handwrtng Recognton Incremental MQDF Learnng for Wrter Adaptve Handwrtng Recognton Ka Dng, Lanwen Jn * School of Electronc and Informaton Engneerng, South

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

Three supervised learning methods on pen digits character recognition dataset

Three supervised learning methods on pen digits character recognition dataset Three supervsed learnng methods on pen dgts character recognton dataset Chrs Flezach Department of Computer Scence and Engneerng Unversty of Calforna, San Dego San Dego, CA 92093 cflezac@cs.ucsd.edu Satoru

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

A Robust Method for Estimating the Fundamental Matrix

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

More information

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

Hybrid Non-Blind Color Image Watermarking

Hybrid Non-Blind Color Image Watermarking Hybrd Non-Blnd Color Image Watermarkng Ms C.N.Sujatha 1, Dr. P. Satyanarayana 2 1 Assocate Professor, Dept. of ECE, SNIST, Yamnampet, Ghatkesar Hyderabad-501301, Telangana 2 Professor, Dept. of ECE, AITS,

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

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Why consder unlabeled samples?. Collectng and labelng large set of samples s costly Gettng recorded speech s free, labelng s tme consumng 2. Classfer could be desgned

More information

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

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

More information

Real-Time View Recognition and Event Detection for Sports Video

Real-Time View Recognition and Event Detection for Sports Video Real-Tme Vew Recognton and Event Detecton for Sports Vdeo Authors: D Zhong and Shh-Fu Chang {dzhong, sfchang@ee.columba.edu} Department of Electrcal Engneerng, Columba Unversty For specal ssue on Multmeda

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

Local Quaternary Patterns and Feature Local Quaternary Patterns

Local Quaternary Patterns and Feature Local Quaternary Patterns Local Quaternary Patterns and Feature Local Quaternary Patterns Jayu Gu and Chengjun Lu The Department of Computer Scence, New Jersey Insttute of Technology, Newark, NJ 0102, USA Abstract - Ths paper presents

More information

An efficient method to build panoramic image mosaics

An efficient method to build panoramic image mosaics An effcent method to buld panoramc mage mosacs Pattern Recognton Letters vol. 4 003 Dae-Hyun Km Yong-In Yoon Jong-Soo Cho School of Electrcal Engneerng and Computer Scence Kyungpook Natonal Unv. Abstract

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

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

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

The Research of Support Vector Machine in Agricultural Data Classification

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

More information

A Probabilistic Approach to Detect Urban Regions from Remotely Sensed Images Based on Combination of Local Features

A Probabilistic Approach to Detect Urban Regions from Remotely Sensed Images Based on Combination of Local Features A Probablstc Approach to Detect Urban Regons from Remotely Sensed Images Based on Combnaton of Local Features Berl Sırmaçek German Aerospace Center (DLR) Remote Sensng Technology Insttute Weßlng, 82234,

More information

Wishing you all a Total Quality New Year!

Wishing you all a Total Quality New Year! Total Qualty Management and Sx Sgma Post Graduate Program 214-15 Sesson 4 Vnay Kumar Kalakband Assstant Professor Operatons & Systems Area 1 Wshng you all a Total Qualty New Year! Hope you acheve Sx sgma

More information

Using Fuzzy Logic to Enhance the Large Size Remote Sensing Images

Using Fuzzy Logic to Enhance the Large Size Remote Sensing Images Internatonal Journal of Informaton and Electroncs Engneerng Vol. 5 No. 6 November 015 Usng Fuzzy Logc to Enhance the Large Sze Remote Sensng Images Trung Nguyen Tu Huy Ngo Hoang and Thoa Vu Van Abstract

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

An optimized workflow for coherent noise attenuation in time-lapse processing

An optimized workflow for coherent noise attenuation in time-lapse processing An optmzed workflow for coherent nose attenuaton n tme-lapse processng Adel Khall 1, Hennng Hoeber 1, Steve Campbell 2, Mark Ibram 2 and Dan Daves 2. 1 CGGVertas, 2 BP 75 th EAGE Conference & Exhbton ncorporatng

More information

TECHNIQUE OF FORMATION HOMOGENEOUS SAMPLE SAME OBJECTS. Muradaliyev A.Z.

TECHNIQUE OF FORMATION HOMOGENEOUS SAMPLE SAME OBJECTS. Muradaliyev A.Z. TECHNIQUE OF FORMATION HOMOGENEOUS SAMPLE SAME OBJECTS Muradalyev AZ Azerbajan Scentfc-Research and Desgn-Prospectng Insttute of Energetc AZ1012, Ave HZardab-94 E-mal:aydn_murad@yahoocom Importance of

More information

Integrated Expression-Invariant Face Recognition with Constrained Optical Flow

Integrated Expression-Invariant Face Recognition with Constrained Optical Flow Integrated Expresson-Invarant Face Recognton wth Constraned Optcal Flow Chao-Kue Hseh, Shang-Hong La 2, and Yung-Chang Chen Department of Electrcal Engneerng, Natonal Tsng Hua Unversty, Tawan 2 Department

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

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

Positive Semi-definite Programming Localization in Wireless Sensor Networks

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

More information

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

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

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

EXTENDED BIC CRITERION FOR MODEL SELECTION

EXTENDED BIC CRITERION FOR MODEL SELECTION IDIAP RESEARCH REPORT EXTEDED BIC CRITERIO FOR ODEL SELECTIO Itshak Lapdot Andrew orrs IDIAP-RR-0-4 Dalle olle Insttute for Perceptual Artfcal Intellgence P.O.Box 59 artgny Valas Swtzerland phone +4 7

More information

Biostatistics 615/815

Biostatistics 615/815 The E-M Algorthm Bostatstcs 615/815 Lecture 17 Last Lecture: The Smplex Method General method for optmzaton Makes few assumptons about functon Crawls towards mnmum Some recommendatons Multple startng ponts

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

USING LINEAR REGRESSION FOR THE AUTOMATION OF SUPERVISED CLASSIFICATION IN MULTITEMPORAL IMAGES

USING LINEAR REGRESSION FOR THE AUTOMATION OF SUPERVISED CLASSIFICATION IN MULTITEMPORAL IMAGES USING LINEAR REGRESSION FOR THE AUTOMATION OF SUPERVISED CLASSIFICATION IN MULTITEMPORAL IMAGES 1 Fetosa, R.Q., 2 Merelles, M.S.P., 3 Blos, P. A. 1,3 Dept. of Electrcal Engneerng ; Catholc Unversty of

More information

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

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

More information

A New Feature of Uniformity of Image Texture Directions Coinciding with the Human Eyes Perception 1

A New Feature of Uniformity of Image Texture Directions Coinciding with the Human Eyes Perception 1 A New Feature of Unformty of Image Texture Drectons Concdng wth the Human Eyes Percepton Xng-Jan He, De-Shuang Huang, Yue Zhang, Tat-Mng Lo 2, and Mchael R. Lyu 3 Intellgent Computng Lab, Insttute of Intellgent

More information

Novel Pattern-based Fingerprint Recognition Technique Using 2D Wavelet Decomposition

Novel Pattern-based Fingerprint Recognition Technique Using 2D Wavelet Decomposition Mathematcal Methods for Informaton Scence and Economcs Novel Pattern-based Fngerprnt Recognton Technque Usng D Wavelet Decomposton TUDOR BARBU Insttute of Computer Scence of the Romanan Academy T. Codrescu,,

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

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

An Entropy-Based Approach to Integrated Information Needs Assessment

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

More information

PERFORMANCE EVALUATION FOR SCENE MATCHING ALGORITHMS BY SVM

PERFORMANCE EVALUATION FOR SCENE MATCHING ALGORITHMS BY SVM PERFORMACE EVALUAIO FOR SCEE MACHIG ALGORIHMS BY SVM Zhaohu Yang a, b, *, Yngyng Chen a, Shaomng Zhang a a he Research Center of Remote Sensng and Geomatc, ongj Unversty, Shangha 200092, Chna - yzhac@63.com

More information

Long-Term Moving Object Segmentation and Tracking Using Spatio-Temporal Consistency

Long-Term Moving Object Segmentation and Tracking Using Spatio-Temporal Consistency Long-Term Movng Obect Segmentaton Trackng Usng Spato-Temporal Consstency D Zhong Shh-Fu Chang {dzhong, sfchang}@ee.columba.edu Department of Electrcal Engneerng, Columba Unversty, NY, USA Abstract The

More information

Research and Application of Fingerprint Recognition Based on MATLAB

Research and Application of Fingerprint Recognition Based on MATLAB Send Orders for Reprnts to reprnts@benthamscence.ae The Open Automaton and Control Systems Journal, 205, 7, 07-07 Open Access Research and Applcaton of Fngerprnt Recognton Based on MATLAB Nng Lu* Department

More information

Robust visual tracking based on Informative random fern

Robust visual tracking based on Informative random fern 5th Internatonal Conference on Computer Scences and Automaton Engneerng (ICCSAE 205) Robust vsual trackng based on Informatve random fern Hao Dong, a, Ru Wang, b School of Instrumentaton Scence and Opto-electroncs

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

Feature Selection for Target Detection in SAR Images

Feature Selection for Target Detection in SAR Images Feature Selecton for Detecton n SAR Images Br Bhanu, Yngqang Ln and Shqn Wang Center for Research n Intellgent Systems Unversty of Calforna, Rversde, CA 95, USA Abstract A genetc algorthm (GA) approach

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

Distance Calculation from Single Optical Image

Distance Calculation from Single Optical Image 17 Internatonal Conference on Mathematcs, Modellng and Smulaton Technologes and Applcatons (MMSTA 17) ISBN: 978-1-6595-53-8 Dstance Calculaton from Sngle Optcal Image Xao-yng DUAN 1,, Yang-je WEI 1,,*

More information

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints TPL-ware Dsplacement-drven Detaled Placement Refnement wth Colorng Constrants Tao Ln Iowa State Unversty tln@astate.edu Chrs Chu Iowa State Unversty cnchu@astate.edu BSTRCT To mnmze the effect of process

More information

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

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

Multiple Frame Motion Inference Using Belief Propagation

Multiple Frame Motion Inference Using Belief Propagation Multple Frame Moton Inference Usng Belef Propagaton Jang Gao Janbo Sh The Robotcs Insttute Department of Computer and Informaton Scence Carnege Mellon Unversty Unversty of Pennsylvana Pttsburgh, PA 53

More information

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters Proper Choce of Data Used for the Estmaton of Datum Transformaton Parameters Hakan S. KUTOGLU, Turkey Key words: Coordnate systems; transformaton; estmaton, relablty. SUMMARY Advances n technologes and

More information

Modeling Inter-cluster and Intra-cluster Discrimination Among Triphones

Modeling Inter-cluster and Intra-cluster Discrimination Among Triphones Modelng Inter-cluster and Intra-cluster Dscrmnaton Among Trphones Tom Ko, Bran Mak and Dongpeng Chen Department of Computer Scence and Engneerng The Hong Kong Unversty of Scence and Technology Clear Water

More information

A Modified Median Filter for the Removal of Impulse Noise Based on the Support Vector Machines

A Modified Median Filter for the Removal of Impulse Noise Based on the Support Vector Machines A Modfed Medan Flter for the Removal of Impulse Nose Based on the Support Vector Machnes H. GOMEZ-MORENO, S. MALDONADO-BASCON, F. LOPEZ-FERRERAS, M. UTRILLA- MANSO AND P. GIL-JIMENEZ Departamento de Teoría

More information

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

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

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

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

Face Recognition Based on SVM and 2DPCA

Face Recognition Based on SVM and 2DPCA Vol. 4, o. 3, September, 2011 Face Recognton Based on SVM and 2DPCA Tha Hoang Le, Len Bu Faculty of Informaton Technology, HCMC Unversty of Scence Faculty of Informaton Scences and Engneerng, Unversty

More information

Audio Content Classification Method Research Based on Two-step Strategy

Audio Content Classification Method Research Based on Two-step Strategy (IJACSA) Internatonal Journal of Advanced Computer Scence and Applcatons, Audo Content Classfcaton Method Research Based on Two-step Strategy Sume Lang Department of Computer Scence and Technology Chongqng

More information

Combining Multiple Resources, Evidence and Criteria for Genomic Information Retrieval

Combining Multiple Resources, Evidence and Criteria for Genomic Information Retrieval Combnng Multple Resources, Evdence and Crtera for Genomc Informaton Retreval Luo S 1, Je Lu 2 and Jame Callan 2 1 Department of Computer Scence, Purdue Unversty, West Lafayette, IN 47907, USA ls@cs.purdue.edu

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

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