1. The MPEG-7 Visual Standard for Content Description-An Overview Thomas Sikora, Senior Member, IEEE
A presentation by Modupe Omueti
For
CMPT 820:Multimedia Systems
Spring 2005

2. Contents
Introduction
Scope
Methodology
Visual Descriptors
Conclusion

3. Introduction Moving Pictures Expert Group
MPEG-1 for interactive video (1992)
MPEG-2 for digital television (1994)
MPEG-4 for multimedia with emphasis on visual objects (1998 v1, 1999 v2)
MPEG-7 for multimedia content description (2001)

4. Trends Initially few sources of audio, image and video
Increase in volume of digitized audio, images and video
Still images digital video

5. MPEG-7 Formally named Multimedia Content Description Interface
Supports some degree of interpretation of the information’s meaning
Interpretation can be passed on to or accessed by a device or computer code
Not aimed at one application in particular

6. Scope Goals Elements Figure 1: Scope of MPEG-7
Standardized descriptions
Meaningful descriptions
Elements
Description tools: visual decriptors and description schemes
Description Definition Language
System tools
Figure 1: Scope of MPEG-7
Figure 2: MPEG-7 main elements

7. Normative part of MPEG-7 standard
Figure 1: Scope of MPEG-7

8. Figure 2: MPEG-7 main elements

9. Applications Digital libraries (image catalogue, film)
Broadcast media selection (TV channels)
Investigation services (human characteristics recognition, forensics)
Multimedia editing (personalized electronic news service)
Figure 3: Abstract Representation

10. Figure 3: Abstract representation of possible applications using MPEG-7

11. Methodology Standard Development
Specification for Technology Requirements
Technology Request
Proposal Evaluation
Experimentation Model Definition
Core Experiments

12. Visual Descriptors General visual descriptors Domain specific
Color, texture, shape, and motion features
Domain specific
Identification of human faces and face recognition

13. Visual Color Descriptors
Color Spaces (HSV, HMMD)
Supports above for normative purposes
Also supports RGB, YCbCr color spaces
Scalable color descriptor Figure 4
Global color Distribution of Images in color histograms
HSV space, uniformly quantized into 255 bins
Haar Transform used to encode histogram
Histogram bin non-uniformly quantized
color coefficients or histogram bin values for matching

14. Visual Color Descriptors
Dominant color descriptor
Global + local spatial color distribution
Colors clustered into a small no of representative colors
representative color, %age, spatial coherency, variance
Color layout descriptor
Spatial distribution of color in an arbitrarily shaped region
Color structure descriptor
HMMD, local color feature, sliding window
Histogram on color appearance count
Group of Frames/Group of Pictures
SCD for a collection of similar images (frames) or video frames
Average, median, intersection histograms of GoF or GoP

15. Figure 4: Three color images and their MPEG-7 histogram color distribution, depicted using a simplified color histogram. Based on the color distribution, the two left images would be recognized as more similar compared to the one on the right.

16. Visual Texture Descriptors
Texture Features
Visual patterns (homogenous or non-homogenous)
Multiple colors in images
Multiple intensities in images
Surface structural information Figure 5

17. Figure 5: Examples of grayscale images with different textures
Figure 5: Examples of grayscale images with different textures. Using the MPEG-7 Visual texture descriptors, the two images on the bottom would be rated of similar texture, while less similar in texture compared to the two images on the top.

18. Visual Texture Descriptors
Homogenous texture descriptor Figure 6
Scale and orientation sensitive filters
Mean and SD of frequency coefficients (RT-FT)
Scale and rotation-invariant description and matching
2D Gabor functions for filtering feature channels
Non homogenous texture descriptor (Edge Histogram)
Spatial distribution of edges
Division of image into 16 non overlapping blocks of equal size
Five edge categories: vertical, horizontal, 45 , 135 , and non directional edge.
Rotation-sensitive and rotation-invariant
Non uniform quantization using 3 bits, descriptor size of 240 bits (16x5x3)

19. Figure 6: Frequency layout for MPEG-7 Homogenous Texture Descriptor frequency extraction. Energy and energy deviation values are extracted from this frequency division into 30 channels.

20. Visual Shape Descriptors
Provides a powerful visual clue
Invariant to scaling, rotation, and translation
2-D or 3-D in nature
For 2-D there are two categories
Contour based which uses only boundary information of objects
Region-based which the entire shape region

21. Visual Shape Descriptors
3-D Shape Descriptor—Shape Spectrum
Based on a shape spectrum concept
Histogram of a shape index
Measures local convexity of each local 3-D surface
Histograms with 100 bins are used—each quantized by 12 bits.
Region Based Shape Descriptor (Art) Figure 7
Uses all pixels constituting a shape within a frame
Region-based moments invariant to transformations
Coefficients of ART basis functions quantized

22. Figure 7: Examples of various shapes that can be indexed using MPEG-7 Region-Based Shape Descriptor. Images contained in either of the sets (a)–(d) would be rated similar and dissimilar to the ones in the remaining sets. For example, images in set (a) would be identified being similar and dissimilar to the ones in set (b), (c), or (d).

23. Types of Visual Shape Descriptors
Contour based shape descriptor Figure 9
Curvature scale-space (CCS)
Eccentricity and circularity values
Robust to non-rigid motion partial occlusion of the shape and perspective transformations
2-D/3-D shape descriptor
Representation of 3-D objects using multiple 2-D snapshots

24. Figure 8: Examples of shapes that can be indexed using MPEG-7 Contour-Based Shape Descriptor.

25. Motion Descriptors for Video
Motion Activity Descriptors
Activity level and pace of motion in a scene
Motion activity intensity descriptor
SD of motion vector magnitude
SDs quantized into five activity levels
Optional Features
motion direction
spatial distribution of motion activity
Temporal distribution of motion activity
Camera Motion Descriptor Figure 9
Global motion parameters in time
zoom activity
translatory motion
Motion similarity matching in particular time periods

26. Figure 9: Camera model for MPEG-7 Camera Motion Descriptor
Figure 9: Camera model for MPEG-7 Camera Motion Descriptor. Perspective projection to image plane p and camera motion parameters. The (virtual) camera is located in O.

27. Motion Descriptors for Video
Warping Parameters
Parametric motion descriptor
Object description using 2-D parametric models
translations, rotations, scaling and combination of them
planar perspective models
quadratic models
Arbitrary objects, defined as regions (group of pixels) in the image over a specified time interval
Global sprite or mosaic
Motion Trajectory
Description for independently moving objects
Object displacement over time

28. Conclusion Identify, filter and browse images using visual content
Specification to allow interoperability and flexibility
Other MPEG-7 standards
Storage, access and transmission of descriptors and descriptors schemes in system specification

30. Thank you