WWW.DISSERS.RU


...
    !

Pages:     | 1 |   ...   | 20 | 21 || 23 | 24 |   ...   | 54 |

4.5. Concept Entity Class The cognitive and affective content of an actor is modeled as a concept object. The concept is modeled as a separate entity type because of its ontological nature, thereby improving the semantic search. Formally, a concept entity can be defined as Knowledge Engineering Concept = { CID, AID, EID, T, D } where T = { Emotion, Feeling, Mood } and D denote type of the concept and description as a string using natural language respectively.

4.6. Interaction Relationships An interaction relationship relates members of an entity set to those of one or more entity sets. The DVSM employs the following set of relationships between the different entity sets. They are Composition (C), Spatial(S):

which are topological (Egenhofer, 1994) and directional (Li, 1996), Temporal(T): Allens interval algebra (Allan, 1983), Spatio-temporal, Motion(M): such as approach, diverge, stationary which are defined over the basic temporal relations (Athanasios, 2005), Semantic(SE) and Ontological(O).

The following section describes the set of relationships that occur between the various dance video entity sets.

4.6.1. Event Relationship The relations between events are composition and temporal. Intuitively, a dance step of an actor may be followed by another actor immediately. Similarly, a dance step of an actor may be repeated by another dancer some time later. These follows and repeats relations are cues for later retrieval and mining operations. For example the query, find the set of dance steps done by a dancer, that is repeated by another dancer, can be processed by checking the life spans of the corresponding events.

Suppose E1, E2,..., En are dance events participating in a temporal relationship. Let a1 and a2 be the actors, xand x2 be the actions of agents present in E1 and E2 respectively. Then, the predicate REPEATS:S(X)S(A)E can take an actor and action and can return a set of events in which the action is performed. There is a constraint on the REPEATS predicate.

Constraint.2. Let LS1 and LS2 be the lifespan of E1 and E2 respectively. Then (x1 = x2 ) V (LS1 < LS1) > (E1 = E2) Similarly, the other predicates such as performSameStep, performDifferentStep, and observe can be formulated, apart from follows and repeats predicates. Event relationships are formally defined as follows:

EE = { SRC, TAR, LST } where SRC and TAR denote the source and target event ids and LST is the set of composition and temporal relationships which hold between source and target events.

4.6.2. Object Relationship Objects can be composed of other objects. For example, consider Figure2 where hero holds a flower in his right hand. Here, flower is an example of an object. Formally, the relationship between objects can be represented similar to event relationships with a restriction that the SRC and TAR can be basic entities and LST will contain only composition relations.

4.6.3. Actor Relationship Actor relationship represents the relationship between the roles of the objects, such as relation between hero and heroine who are dancer objects. Spatial, temporal and semantic relationships exist between the actors in a particular dance event. For instance, hero standing left to the heroine initially, may approach the heroine. This dance semantic contains spatial and motion relationships left and approach respectively. The actor relationship is formally defined as shown below:

AA = { AID1, AID2, O1, O2, LST } where AID1 and AID2 are roles of the dancers O1 and O2 respectively and LST is now the set containing spatial, temporal and semantic relationships. Note that O1 and O2 are basic entity types.

Fourth International Conference I.TECH 2006 4.6.4. Agent Relationship Agent relationship is a second level semantic relation that describes the spatial and temporal characteristics of the agents. That is, agent relationship represents the finer semantics between the body parts of an actor. For instance, heroine is touching her left cheek with the index finger of her right hand. So, left cheek and right hand are the agents and finger can be the instrument used in the semantic relationship touch. Agent relationship is formally defined as, AGAG = { AGID1, AGID2, AID, LST } where AGID1 and AGID2 are agentIDs of an actor AID and LST is similar to actor relationships.

4.6.5. Concept Relationship Concept relationship is an ontological relationship (O) between concept entities. Typical ontological relationships (Guiness, 2004) are subClassOf, cardinality, intersection and union. This relationship is similar to event relationship with a modification that the source and target ids represent concepts and LST holds only ontological relations.

All other types of relationships between the different dance video entities are either semantic relationships or composition relationships such as partOf, composedOf, memberOf and so on. Table 1 summarizes the semantics of the kinds of relationships that exist between the dance video entities.

Table 1. Semantics of Relationships.

Event Object Actor Agent Concept Event C,T,SE C C Object C C C Actor C C S,T,SE C C Agent C C S,T,SE Concept C C O 5. Implementation of DVSM We have implemented the model in order to annotate the macro and micro features that are associated with the dance video. The tool has been developed using J2SE1.5 and JMF2.1.1 under Dell workstation. The tool is interactive as it minimizes the hard coding.

The dance video can be annotated by looking at the video clips that is running. Macro features can be annotated initially. The details of the dancers, musician, music, song, background, tempo, dance origin, context (whether live, rehearsal, professional play, competition etc), date and time of recording, type of performance venue and type of dance video are annotated. The screen shot depicting the rendering of the dance and interactive annotation of macro features is shown in Figures 3.

Then, micro features of every dance step of a song have to be annotated. The screen shot depicted in Figure represents events, actors, agents and concepts. The annotator, by looking at the video, annotates the different information pertaining to these entity types in the order: event, actors of this event, agents of the actors, concepts revealed by the actors. But, one can swap the annotation of agents and concepts depending on his interest.

The user interface has been carefully designed such a way that it minimizes the hard coding, as many of the graphical components will be populated automatically.

The second part of the micro features annotation involves the description of the various relationships between the entity types. For instance, event relationships, actor relationships, agent relationships and concept relationships describe the spatial, temporal, motion and semantic relations that exist between the entity types. The annotated data are stored in a backend database.

Knowledge Engineering Fig.3 and 4. Screen shot of macro and micro annotation of objects.

6. Evaluation Batini(1996) posits that conceptual model should possess the basic qualities: expressiveness, simplicity, minimality and formality. Additionally, Harry(2001) outlines other semantic qualities for video types. They are:

explicit media structure(M), ability to describe objects (O), events (E), spatial relationships(S), temporal relationships(T) and integration of syntactic and semantic information (I). This paper introduces another factor, contextual description (Actor(A) and Agent(G)) to evaluate the proposed model.

The DVSM satisfies all the semantic quality requirements. Moreover, DVSM is unique in modeling the finer spatio-temporal contextual semantics of the events at finer granularity, with the help of agent entity type. Table contrasts the existing semantic content based models against the semantic quality factors. The table illustrates that some models lack semantic features, some lack syntactic features and only few models integrate both syntactic and semantic relationships. Some applications, like soccer sports video, require the model to represent the contextual features of objects (called actors). The ESSVM proposed by Ahmet et al, describes contextual information at actor level. However, dance videos require contextual description at multiple granularities (called agent), beyond the actor level. Our proposed semantic model possesses both contextual abstractions-actor and agent, apart from the other semantic qualities. Hence, with the agent based approach, the paper claims to have achieved conceptual, semantic and contextual qualities in dance video data modeling.

Table 2. Comparison of semantic video models.

Legend: M-Media structure, O-Object, E-Event, S-Spatial, T-Temporal, I-Syntactic, semantic information, A-Actor, G-Agent.

Model Semantic Qualities M O E S T I A G AVIS OVID QBIC DISIMA COSMOS7 ATN ESSVM DVSM Fourth International Conference I.TECH 2006 7. Conclusion Data semantics provides a connection from a database to the real world outside the database and the conceptual model provides a mechanism to capture the data semantics (Vijay, 2004). The task of conceptual modeling is crucial and important, because of the vast amount of semantics that exist in multimedia applications. In particular, dance videos possess several interesting semantics for modeling and mining. This paper described as agent based approach for elicitation of the semantics such as macro and micro features of the dance videos. An interactive annotation tool has been developed based on the DVSM for annotating the dance video semantics at syntactic, semantic and contextual levels. Since dance steps are annotated manually, it is somewhat tedious to annotate dances by the dance expert.

Further work would be useful in many areas. It would be interesting to explore how DVSM can be used as a video model for exact and approximate query processing. As MPEG-7 is used to document the video semantics recently, it is valuable to employ MPEG-7 for representing dance semantics to enable better interoperability.

Finally, it will be useful to explore how video mining techniques can be applied to dance videos.

Acknowledgements This work is supported fully by the University Grants Commission (UGC), Government of India grant XTFTNBD065. The authors thank the editors and the anonymous reviewers for their insightful comments.

Bibliography Adali, S, Candan, K.S, Chen, S.S, Erol,K & Subramanian, V.S.(1996). The Advanced Video Information System: Database structures and query processing. Multimedia Systems. 4, 172-186.

Ahmet Ekin, Murat Tekalp, A & Rajiv Mahrotra.(2004). Integrated semantic-syntactic video event modeling for search and browsing. IEEE Transaction on Multimedia, 6(6), 839-851.

Allen, J.F.(1983). Maintaining knowledge about temporal intervals. Communication of ACM, 26(11), 832-843.

Al Safadi, L.A.E & Getta, J.R.(2000). Semantic modeling for video content based retrieval systems. 23rd Austral Asian Computer Science conference, 2-9.

Ann Hutchinson, G.(1984). Dance Notation: Process of recording movement. London: Dance Books.

Antani, S, Kasturi, R & Jain, R.(2002). A survey on the use of Pattern recognition methods for abstraction, indexing and retrieval of images and video. Pattern Recognition, 35(4), 945-965.

Athanasios.C, Gkoritsas, & Marios.C.Angelides.(2005). COSMOS-7: A video content modeling framework for MPEG-7.

MMM05, 123-130.

Batini, C, Ceri, S & Navathe, S.B.(1992). Conceptual database design: An entity relationship approach, Benjamin/Cummings publishing company.

Chen, P.P.(1976). The Entity-Relationship model-Towards a unified view of data. ACM Transaction on Database Systems, 1(1), 9-36.

Cheng Youg & XU.De.(2003). Hierarchical semantic associative video model. IEEE conference NNSP03, 1217-1220.

Colombo, C, Del,A, Bimbo & Pala, P.(1999). Semantics in visual information retrieval. IEEE Multimedia, 6(3),.38-53.

Dorai,C, Manthe,A, Nack,F, Rutledge,L, Sikora,T & Zettl,H.(2002). Media Semantics: Who needs it and why. ACM conf.

Multimedia, 580-583.

Egenhofer, M & Franzosa, R.(1994). Point-set topological spatial relations, Conference of GIS, 5(2),161-174.

Forouzan Golshani, Pegge Vissicaro & Yound Choon Park.(2004). A multimedia information repository for cross cultural dance studies. Multimedia Tools and Applications, 24, 89-103.

Guiness, D.M, Van, F & Harmelen(2004). OWL: Web Ontology Language Overview. W3C Recommentation, http:// www.w3.org/tr/owl-features/.

Harry.W.Agius & Marios.C.Angelides(2001). Modeling content for sematic level querying of multimedia. Multimedia Tools and Applications, 15, 5-37.

Hutchinson, A.(1954).Labanotation: A system for recording movement. London: T Art Books.

Koh, J.L, Lee,C.S &.Chen, A.L.P.(1999). Semantic video model for content based retrieval. IEEE Conference MMCS99, 2, 472-478.

Knowledge Engineering Lei Chen & Tamer Ozsu, M.(2003). Modeling video data for content based queries: Extending DISIMA image data model.

MMM'03, 169-189.

Li, J.Z, Ozsu, M.T & Szafron, D.(1996). Modeling of video spatial relationships in a object oriented database management systems. IEEE Workshop on MMDMS, 124.

Martinez, J.M.(2003). MPEG-7 Overview, ISO/IEC JTC1/SC29/WG11-N4980.

Oomoto,E,& Tanaka, K.(1993). OVID: Design and implementation of a video object database system. IEEE Transaction on.

Knowledge and Data Engineering, 5(4), 629-643.

Saraswathi.(1994). Bharatha Nattiya Kalai.Chennai: Thirumagal Nilayam.

Shu Ching Chen, Mei Ling Shyu, R.L.Kashyap.(2000). ATN as a semantic model for video data. Journal of Network Information Management, 3(1), 9-25.

Smeulders, A.W.M, Worring, M, Santini, S, Gupta, A &.Jain, R.(2000). Content based image retrieval at the end of early days. IEEE Transaction on. Pattern Analysis and Machine Intelligence. 22(12),.1349-1380.

Vendrig, J.W.(2002). Interactive adaptive movie annotation, Multimedia and Expo conference, 1, 93-96.

Vijay Khatri, Sudha Ram & Richard.T.Snodgrass.(2004). Augmenting a conceptual model with geospatiotemporal annotations. IEEE Transaction on Knowledge and Data Engineering, 16(11), 1324-1338.

Web of Indian Classical Dances, http:// www.narthaki.com/index.htm.

Pages:     | 1 |   ...   | 20 | 21 || 23 | 24 |   ...   | 54 |



2011 www.dissers.ru -

, .
, , , , 1-2 .