Video Adaptation:
Concept, Technologies, and Open Issues
SHIH-FU CHANG
Presented by Jun-Cheng Chen
03/17/2005
Outline
z
Introduction
z
Unified Conceptual Framework and Technology Taxonomy
z
Active Research Areas
z
Open Issues
z
Support of Adaptation in International Standards
z
Conclusion
Introduction(1/3)
z
Video adaptation:
z Emerging field in pervasive media applications (such as PC, TV, PDA, or cellular phone).
z Transform the input video to an output in video or augmented multimedia.
z utilize manipulations at multiple levels
z Signal, structural, semantic.
z constrained optimization
Introduction(2/3)
z Video adaptation differs from video coding in its scope and intended application locations.
z signal level vs structural level vs semantic level, tanscoding vs selection vs summarization, bandwidth vs power vs time-constrained.
z Often In the intermediate location, such as proxy between server and client.
z Video adaptation is still a relatively less defined field.
z No coherent set of concepts, terminologies, or issues defined over well-formulated problems
Introduction(3/3)
Unified Conceptual Framework
and Technology Taxonomy
Unified Conceptual Framework and Technology Taxonomy
z
Entity:
z Defined to refer to the basic unit of video that undergoes the adaptation process.
z Different levels, such pixel, object, frame, shot, scene, syntactic components, and semantic
components
z Each entity is associated with certain resource requirements and utility values.
Unified Conceptual Framework and Technology Taxonomy
z Utility:
z It represents the quality or users’ satisfaction of the video content (such as PSNR).
z Adaptations space:
z The space of feasible adaptation for a given video entity.
z Different adaptation operators can be defined for different types of entities.
(ex: a video frame can be reduced in resolution, spatial quality, or skipped to reduce bandwidth cost.)
Unified Conceptual Framework and Technology Taxonomy
z
Systematic Procedure for Designing Video Adaptation Technologies
z
Video Adaptation Taxonomy
z Format transcoding
z Selection/Reduction
z Replacement
z Synthesis
Systematic Procedure for Designing Video Adaptation Technologies (1/3)
z
Identify the adequate entities for adaptation.
z
Identify the feasible adaptation operators.
z
Develop models for measuring and
estimating the resource and utility values
associated with video entities undergoing
identified operators.
Systematic Procedure for Designing Video Adaptation Technologies(2/3)
z
Given user preferences and constraints on resource or utility, develop strategies to find the optimal adaptation operator(s) satisfying the constraints.
Problem formulation:
Given a content entity E, user preferences, and resource
constraints Cr, find the optimal adaptation operations Aopt within the feasible adaptation region so that the utility of the adapted entity e’ is maximized.
Systematic Procedure for Designing Video Adaptation Technologies(3/3)
[10].Y. Wang, J.-G. Kim, and S.-F. Chang, “Content-based utility function prediction for real-time MPEG-4 transcoding,” presented at the
IEEE Int. Conf. Image Processing, Barcelona, Spain, 2003.
Video Adaptation Taxonomy
z
Format transcoding:
z To transcode video from one format to another, in order to make the video compatible with the new usage environment.
z
Selection/Reduction:
z Select some components of the entity and reduce them for saving resources.
z Example: We can change the bit rate, frame rate or resolution for shots and frames in a video clip,
Video Adaptation Taxonomy
z Replacement:
z Replace selected elements in a video entity with less expensive counterparts, while aiming at preserving the overall perceived utility.
z Example: a video sequence may be replaced with key frames.
z Synthesis:
z Synthesize new content presentations based on analysis results.
Video Adaptation Taxonomy
Active Research Areas
z
Semantic Event-Based Adaptation
z
Structural-Level Adaptation
z
Transcoding
z
Rapid Fast-Forward Drastic Temporal
Condensation
Semantic Event-Based Adaptation
z
Doing video analysis for events and boundaries detection.
z By using the information of video content, such as
“the scoring points in sports video”, and “the breaking news in broadcast programs”.
z
Results of video event analysis can be
utilized to produce different forms of
adaptation.
Semantic Event-Based
Adaptation
Semantic Event-Based Adaptation
z
Example: The percentage of important segments in the whole stream (such as sports broadcast).
z They found non-important segments occupy more than 50% of duration.
z Their system which focuses on sports can reach higher than 90% accuracy
[6]S.-F. Chang, D. Zhong, and R. Kumar, “Real-time content-based adaptive streaming of sports video,” presented at the IEEE Workshop
Structural-Level Adaptation
z
The structures in video are caused by event occurrence orders, camera control patterns, and the final editing process.
z
Exploration of relations of structural elements provides great potential for video adaptation.
z
Example:
z Key frame extraction
z Mosaicing
Structural-Level Adaptation
Transcoding
z
Signal level adaptation
z
Involving various manipulations of coded representations and issues of bit allocation
z
Manipulation of video signals:
z Spatial: change spatial resolution, i.e., frame size.
z Precision: change the bit plane depth, color depth, or the step size for quantizing the transform
coefficients.
Rapid Fast-Forward Drastic Temporal Condensation
z Video skimming
z Bad ways:
z Increase the frame rate of the player.
z Make the audio track unrecognizable.
z Uniformly sample the frames in the original sequence.
z Important video frames may be skipped and audio content may be unrecognizable.
z Extract keyframes to form shorter image sequence.
z Lose the synchronization between video and the associated
Rapid Fast-Forward Drastic Temporal Condensation
[14].H. Sundaram, L. Xie, and S.-F. Chang, “A utility framework for the automatic generation of audio-visual skims,”
presented at the ACMMultimedia Conf., Juan Les Pins, France, 2002.
z Adaptation entities: video shots.
z Adaptation operations: length trimming or dropping of individual shots.
z The problem was formulated as constrained optimization.
Constraints: viewing time, dialogs, key phrases, key audio, etc.
Open Issues
z
Define Utility Measures and User Preferences
z
Resolve Ambiguity in Specifying Adaptation Operation
z
Relations Among Adaptation, Utility, and Resource
z
Search Optimal Solutions in Large Spaces
Define Utility Measures and User Preferences
z It is difficult to define a universal measure for different levels or dimensions.
z Levels include Perceptual, semantic, and comprehensiveness.
z Signal-level measures are often inadequate m
z many high-level operations such as shot removal, modality replacement, etc.
z These operations also cause complex changes to content at other levels.
z Users preferences often vary with content, task, and usage environment.
Define Utility Measures and User Preferences
z
Some possible alternatives
z Infer user preferences based on the usage history.
z Correlate subjective preferences with content characteristics.
Resolve Ambiguity in Specifying Adaptation Operation
z Some adaptation operations are not unambiguously defined.
z “remove the second half of each shot”
z “drop 10 % of transform coefficients”
z Some possible ways
z Restrict adaptation operation only on unambiguous
representation formats, such as JPEG 2000 and MPEG-4 fine grained scalable schemes.
z Estimate the bound of variations in resource and utility.
Relation Among Adaptation, Utility, and Resource
z Relations among adaptation, resource, and utility are often complex.
z The complexity is especially high when the dimensionality of each space is high.
z Potential approaches
z Sample the adaptation space and store the corresponding resource and utility values.
z Decompose the adaptation space into low-dimensional spaces and sample each subspace separately.
Search Optimal Solution in Large Spaces
z Exploration of the above multi-space relations often leads to formulation of constrained optimization.
z Analytical solutions may exist for some cases.
z example: rate-distortion model (low dimensional cases)
z Adaptation space: quantization
z Resource space: bit rate
z Utility space: SNR
z In general, each space may have high
dimensionality and the relations across spaces may be complex.
Design End-to-End Integrated Systems
z
Difficulties
z Require joint consider joint consideration of the adaptation subsystem with other subsystems.
z Inconsistent and imperfect content analysis subsystem
z Rights management
z Content owners impose many restrictions on video content altering.
Design End-to-End Integrated Systems
z
Possible solutions:
z Adopt modular designs of subsystems and
provide well-defined abstraction of requirements and performance of each subsystem.
z Follow the international standard which are needed for describing information related to media rights management.
Support of Adaptation in International Standards
z
Mpeg-7 Content Descriptions
z
Mpeg-21 Digital Item Adaptation
z
Standardized Adaptation Framework
Mpeg-7 Content Descriptions
z Descriptors (Ds) & Description schemes (DSs)
z XML
z Usage history DS
z UserPreferences DS (creators, time periods, locations, etc.)
z Summary descriptions
z Variation descriptions
z Transcoding hints
z Motion hints (for guiding motion-based transcoding methods)
z Semantic importance hints (for guiding rate control)
z Etc…
Mpeg-21 Digital Item Adaptation
z Digital Item Adaptation(MPEG21 part7):
z Address an extended scope of issues related to adaptation of digital multimedia content.
z Usage environment descriptions (UEDs)
z Used to describe a wide array of user, terminal capabilities, network, and natural environment characteristics.
z Universal constraints description (UCD) tool
z similar to UEDs
z more explicit
z AdaptationQos tool
Standardized Adaptation
Framework
Conclusion
z
Despite the burgeoning activities and
advances, this field is in need of an analytical foundation and solutions to many challenging open issues.
z
