DEWS2008 E4-1, 152-8552 2-12-1 ( ) 112-0002 1-1-17 152-8552 2-12-1 152-8550 2-12-1 E-mail: {hanhlh,tetsu}@de.cs.titech.ac.jp, thitiporn.lertrusdachakul@nts.ricoh.co.jp, yokota@cs.titech.ac.jp E- MPMeister e- Evaluation of Methods to Extract Important Scenes for Automatic Digest Generation from a Presentation Video Le HIEU HANH, Thitiporn LERTRUSDACHAKUL, Tetsutaro WATANABE, and Haruo YOKOTA, Department of Computer Science, Faculty of Engineering, Tokyo Institute of Technology Ookayama 2-12-1, Meguro-ku, Tokyo, 152-8552 Japan Application Lab, Software R&D Group, Ricoh Company, Ltd. Tomin-Nissei-Kasugacho Bldg.Koishikawa 1-1-17, Bunkyo-ku, Tokyo, 112-0002 Japan Department of Computer Science, Graduate School of Information Science and Engineering, Tokyo Institute of Technology Ookayama 2-12-1, Meguro-ku, Tokyo, 152-8552 Japan Global Scientific Information and Computing Center, Tokyo Institute of Technology Ookayama 2-12-1, Meguro-ku, Tokyo, 152-8550 Japan E-mail: {hanhlh,tetsu}@de.cs.titech.ac.jp, thitiporn.lertrusdachakul@nts.ricoh.co.jp, yokota@cs.titech.ac.jp Abstract In the recent years, the use of presentation video in E-learning is increasing. On the user s point of view, this often requires previewing the digest of such video before watching the original one. This paper proposes a number of methods of extracting important scenes for automatic digest generation digest from the presentation video which is recorded by MPMeister II - the tool for Multimedia Web Contents. The important scenes extractions are based on several factors, such as word frequency, specificity, duration and orders of the slides. Finally, the results of each methods are evaluated by comparing with the answer sets selected by testers. Key words e-learning, information extraction, video digest
%'&)(+*',!!! #"! #$ 1 1. E- ( ) MPMeister [1] UPRISE [2] ( 1 ) MPMeister ( ) 2. 3. MPMeister 4. 5. 6. 2. 2. 1 Sanderson [3] Yihong [4] Barzilay [5] 2. 2 [6] 2. 3 Jinjun [7] Marcus J.P ANSES [8] Liwei [9] UPRISE [2] [10] [11] [12] UPRISE UPRISE
BB, + 0 / $"+*, -/.1032 4 "!$#% & '( )!$#% & -. 57685:9<;>=@?@9<A<B BDCFE GIHKJLNMOLNP JLNMOLNPRQS 3. MPMeister 2 MPMeister ( ) Microsoft PowerPoint PowerPoint MPMeister PowerPoint PowerPoint MPMeister MPEG-7 (XML ) MPEG-7 MPEG-7 4. 2 MPMeister MPEG-7 4. 1 3 Sen [13] J-E Ontology [14] 4. 2 4. 2. 1 L l S = {s 1, s 2,..., s N } s i s i = [v i1, v i2,..., v im ] "!$#%'& 3 L 1 (!)#%'* S W = {w 1, w 2,..., w M } v ij W L l L l = [c l1, c l2,..., c ln ] c lk S s i c lk = [v i1, v i2,..., v im ] s i 3 L 1 s 1 c 11 s 2 c 12 L 1 = [c 11, c 12 ], S = {s 1, s 2}, s 1 = [v 11, v 12, v 13, v 14 ], s 2 = [v 21, v 22, v 23, v 24, v 25, v 26 ], v 11 =, v 12 =, v 13 =, v 14 =, v 21 =, v 22 =, v 23 =, v 24 =, v 25 =, v 26 =, W = {,,,,} 4. 2. 2 [15] p(v ij ) p(v ij )
ρ t v ij p(v ij ) = ρ bl v ij b l 0 v ij a ) I d (1) UPRISE [12] I d I d (c lk, θ) = I p (c lk ) t(c lk ) θ (2) t(c lk ) c lk θ I p I p(c lk ) = ( w x c lk c y L l ( v z c y,v z =w x p(v z))) (3) 3 L 1 MPMeister ρ t = 5 ρ bl = 1 c 11 (v 11 ) p(v 11 ) = 5 (v 14 ) p(v 14 ) = 1 I p (c 11 ) = 18, I p (c 12 ) = 25 b ) I dc UPRISE [12] I dc I dc (c lk, θ, δ, ε 1, ε 2 ) = j=γ+δ j=γ δ I d (c lk, θ) E(j γ, ε 1, ε 2 ) (4) δ E(j γ, ε 1, ε 2 ) E(x, ε 1, ε 2) = { exp(ε1x) (x < 0) exp( ε 2x) (x > = 0) θ ε θ = 4, ε 1 = 5.0, ε 2 = 0.5 4 ε 1 > ε 2 c ) I dr idf 1 I d I dc I dr I drc I df I dr I dr (c lk, θ) = w x W app(c lk, w x ) app(c y, w x ) I pl(c lk ) t(c lk ) θ (5) c y L l I pl I pl (c lk ) = p(v x ) (6) v x c lk app(c lk, w x ) c lk w x d ) I drc I drc I drc (c lk, θ, δ, ε 1, ε 2) = j=γ+δ e ) j=γ δ I dr (c lk, θ) E(j γ, ε 1, ε 2) I df (7) I df I df (c lk, θ) = w x W 1 4. 2. 3 1 app(c lk, w x ) I d(c lk, θ) (8) 4. 3
3 F A B F F I d 0.45 0.31 0.37 0.67 0.40 0.50 I dc 0.36 0.24 0.29 0.83 0.42 0.56 I dr 0.45 0.33 0.38 0.50 0.33 0.40 I drc 0.45 0.22 0.30 0.50 0.25 0.33 I df 0.45 0.22 0.30 0.83 0.42 0.56 4 2 A 59 1548 26 85 B 32 1043 33 165 C 31 1674 54 160 D 44 1286 29 110 E 98 4256 43 48 F 67 3306 49 68 Windows Media 4 5. ρ t θ 5. 1 5. 1. 1 A B 2 5. 1. 2 6 6 4 C C F F F = = F = 2 + (2) (4) (5) (7) (8) ρ t = 5, b l ρ bl = 1 θ = 1 δ = 3 ε 1 = 5 ε 2 = 0.5 5. 1. 3 F 3 5 6 5 6 5 6 3 A B F 5 A MPMeister 6 B 32 32 11
>=< ; A@? CB 1230/ 67 54 KK << "" $$ WWW && "" ## (( $$ ** WWWW && ++ )) -- // DDD -- WWW -- KKKK WWWW -- TTT \\\ << -- III :987 ;<=?> = = FFF!! ## %% && )) ++ )) )) FF )) \\ QQ ))!!" "!# #%$ $!& &(' ')*' ) ') ))) + +,' + + ')))) -- ' ')))) -+' '))))... /8 // /0 /1 /2 /3 /4 /5 /6 /7 08 0/ 00 01 02 03 04 05 18 1/ 06 07 10 11 12 13 14 15 16 17 28 2/ 20 21 22 23 24 25 26 27 38 3/ 30 31 32 33 34 35 36 37 99 :::: ;;;; <=> <?@ ABCDE DE GF FFFHHHH = = D= = JJJJ> > > >?? KK?? KK@ @@@ ABC DE < <FH I 5 A! #" "%$ $'& &)( ('* *,+ +.- -0/ /1- -2/43 /435- -2/46 /46 - -2/4673 /46738- -2/:9 ;D ;; ;< ;;=> ;? ;;@A ;B <D ;C <; <<=> <? <<@A <B <C =D =; =<FEHGJI EHGJILK K M7N2O7P0Q7R SSUT7V T7V2W XXXX YYYY ZJ[ X7Y MHNJOHP0Q MJNJOHP QQQ RRRR SUT7V SSS T7V2W T2]J^ X7Y Z7[ ZU[ X2YU\ 6 B 11 4 7 4 I df F 7 6 B "!!$# #$% %$& &(' '*) ),+ +-) ),+/. +/.0) ),+/1 +/12) ),+/13. +/13.4) ),+65 7@ 77 78 779: 7; 77< 7> 8@ 7? 8887 889: 8; 88< 8> 8? 9@ 9 7BADC6E F3GIH6JLK GIH6JLKNMLO3P MLO3PNQLR QSRNTLUIV WDX6TIULV GYHDJDKZMLO3P GYH6JNK MLO[PLQNR QNRNTLUIV TLUIVSF F WLX6T,ULV T,ULVS\ \YQ QL]L^ 7 B () 4 B F ( ) F F I d 0.67 0.40 0.50 0.83 0.45 0.59 I dc 0.83 0.42 0.56 1.00 0.43 0.60 I dr 0.50 0.33 0.40 0.67 0.44 0.53 I drc 0.50 0.25 0.33 0.83 0.38 0.53 I df 0.83 0.42 0.56 0.83 0.42 0.56 5. 2 (1) ρ t θ 5 5. 2. 1 1 C D E F 2 5. 2. 2 1 2 2 ρ t 1 10 1 F F ρ t θ 0 5 0.5 F
8 ρ t F 10 θ 9 θ F 11 θ 5. 2. 3 ρ t θ F 8 9 8 ρ t F 9 F θ = 0 θ F θ θ 10 11 θ > 0 θ θ θ 8 9 I df F 6. 5 3
PowerPoint 18 (DEWS2007) pp.e1-3 2007. [13] Sen http://www.mlab.im.dendai.ac. jp/ yamada/ir/morphologicalanalyzer/sen.html. [14] An Experimental Japanese - English Ontology, http: //www.kanzaki.com/docs/sw/jwebont.html. [15] H. Luhn, A Statistical Approach to Mechanized Encoding and Searching of Literary Information, IBM Journal of Research and Development, vol.1, no.4, pp.309-317, 1957. (19024028) CREST 21 COE [1] Ricoh Japan, MPMeister II Ricoh Japan, http://www. ricoh.co.jp/mpmeister. [2] H. Yokota, T. Kobayashi, T. Muraki, and S. Naoi, UP- RISE: Unified Presentation Slide Retrieval by Impression Search Engine, IEICE Trans. on Info. and Syst., vol.e87- D, no.2, pp.397-406, 2 2004. [3] M.Sanderson, Accurate User Directed Summarization from Existing Tools, Proc. of the 7th International Conference on Information and Knowledge Management (CIKM98), 1998. [4] Y. Gong, and X. Liu, Generic Text Summatization Using Relevance Measure and Latent Semantic Analysis, Proc. of the 24th annual international ACM SIGIR conference on Research and development in information retrieval, pp.19-25, 2001. [5] R. Barzilay, and M. Elhadad, Using Lexical Chains for Text Summarization, Proc. of the Intelligent Scalable Text Summarization Workshop (ISTS 97), 1997. [6] A Study on Statistical Methods for Automatic Speech Summarization Ph.D thesis Tokyo Institute of Technology 2002. [7] J. Wang, C. Xu, E. Chng, and Q. Tian, Sport Highlight Dectection from Keyword Sequences using HMM, Proc. of IEEE International Conference on Multimedia and Expo, ICME 04, vol.1, pp.599-602, 2004. [8] M.J. Pickering, L. Wong, and S.M. Rüer, Image and Video Retrieval, Springer Berlin / Heidelberg, 2003. [9] L. He, E. Snocki, A. Gupta, and J. Grudin, Auto- Summarization of Audio-Video Presentation, Proc. of the seventh ACM international conference on Multimedia (Part 1), pp.489-498, 1999. [10] vol.j88-d-i no.3 pp.715-726 3 2005. [11] 17 (DEWS2006) 2006. [12]