1. Multimedia Database Management System
Wei Tsang Ooi
CS731

2. MMDBMS : Querying Interface, Indexing and Buffer Management

3. Why MMDBMS ? Concurrency Control Recovery Privacy Query Support
Version Control

4. Example of MMDBMS Digital Library News-On-Demand Video-On-Demand
Music Database
Telemedicine
Geographic Information System

5. No Intergration id size fps title filename 000001 530M 30 soam l6.mpg
000002
450M 30
tibor
l7.mpg
000003
600M 30
parag
l5.mpg
000004
510M 30
wei
l4.mpg

6. Semi-intergrated
BLOB
000001b70ae

7. Fully Intergrated   
index  
 
buffers 
storage

8. Nature of Multimedia Data
Large amount of data
Time sensitive
Vague matching

9.  Database Components Query Interface Query Processing storage manager
buffer
manager
index

10. Query Interface and Processing

11. Problems Needs support for : SQL is not suitable.
temporal and spatial relations
“natural” interface
fuzzy query
SQL is not suitable.

12. 13 Temporal Relations Allen 83 starts before meets equals overlaps
during
finished by
finishes
overlapped by
contains
met by
started by
after

13. Spatio Relations Papadias, Theodoridis 96 Topological Relations
Directional Relations

14. Topological Relations
disjoint
meet
overlap
covered by
inside
equal
covers
contains

15. Directional Relations
North NW NE
West
East SW SE
South

16. Spatio-temporal Relations
overlap-after
meet-during
...

17. Querying
Image
Audio
Music
Sound
Speech
Video

18. Querying Image Common approach
allow query by sketches (color, shape, texture) or examples.
perform matching by Feature Vectors F = (v1, v2, ... vn)
e.g. Color Histogram

19. Querying Image Exisiting Systems : QBIC VisualSEEK PhotoBook Virage
FourEyes

20. Querying Music Hawley 93 Input by MIDI Keyboard
Measure relative pitch (U, D, S)
Perform exact match with existing database.

21. Querying Music Ghias, Logan, Chamberlin & Smith 95 Input by humming
Extract relative pitch
Perform approximate matching

22. Querying Music Chou, Chen, & Liu 96 Query by chord
Represents musics by chord
C Am Em F C Am Em F ...
Perform fuzzy match

23. Querying Music Chen & Chen 98 Query by “rhythm” (tempo ?)
Represents musics by rhythm
| ¶¶— | ¶·¶·| ¶¶¶¶ | ...
Perform fuzzy match

24. Querying Sound Wold, Blum, Keisar & Wheaton 96
Analyze audio to extract features
loudness, pitch, brightness, bandwidth and harmonicity
Segment the audio to pieces
Feature Vector Matchings

25. Querying Speech Hauptmann & Witbrock 97 Informedia
Use speech recorgnition to convert audio to text
segment audio using silence detection
Query by speaking keywords

26. Querying Video
Query by speech
Informedia
Visual approach
VideoQ

27. VideoQ Chang et. al 97 User can sketch objects Specifies color texture
shape
motion
duration
camera zoom and pan.

28. VisualQ Example
Titanic Sinking
Someone Skiing

29. Indexing

30. Indexing Requirements support spatio-temporal operations
support fuzzy matches

31. Indexing Images N-dimentional indices for feature vector
Well studied in DB/CG community
Two examples :
VP-tree
R-tree

32. VP-tree R Q P
PQS
VWTU R T S U W V

33. VP-tree R Q P R P U T S S Q T VW U W V

34. R-tree

35. Indexing Audio Audio are modeled as strings Inexact match is needed
Common indices for string search can be used
Example
PAT-tree

36. PAT-tree ab b c
abc c
abc c
ababc
abc
babc bc c

37. Indexing Video Treat time as third dimension
We can use any multidimension indexing structures

38. Buffer Management

39. Buffer Management Minimizing response time
Ensuring continuity and synchronization
Prefetching & Replacement

40. Glossaries
Presentation
Media Stream
Media Object

41. Buffer Management
Relevance Based
Distance Based
Buffer Consumption

42. Relevance Based Moser, Kraiß & Klas 95 : L/MRP Definition:
State = (curr obj, skip)
Relevance : State  Real

43. Relevance Based Prefetch : Replace :
future objects with highest relevance
Replace :
old objects with lowest relevance

44. Example (5,+2)
Relevance
History
Referenced
Skip
Time Scale

45. Distance Based Özden, Rastogi & Silberschatz 96 Definition
distance of a client C is the offset between C and the immediate following client.

46. Distance Based Replace :
blocks consumed by clients with largest distance

47. Example C1 C2 C3 C4

48. Example C1 C2 C3 C4

49. Example C1 C2 C3 C4

50. Buffer Consumption Wu & Yu 97 Definition :
Buffer consumption = amount of buffer used x total time

51. Buffer Consumption Result :
Minimizing buffer consumption increase system throughput
Increase retrieval rate of current stream is better then prefetching next waiting stream

52. Open Problems

53. Querying Query video by action ? Query music by
relative pitch + chord + rhythm ?
Continuing quest to improve accuracy

54. Indexing
Well studied problem
Indexing for approximate string search ?

55. Buffer Management Should take into consideration:
object dependencies (MPEG)