Outline with References
- Overviews: [AIGR96], [KHOS96], [LESK97 ch. 4], [NARA96], [SUBR96], [YU97 ch.11]
- Image Features
- Color Histograms: [GONG96] Use HVC, divide into 11 color zones,
compute overall plus 9 area histograms, code into B* tree
- Shape: [MEHT97] best for trademark retrieval using outline based features
and region based features is: combining moment invariants, UNL Fourier feature
- Spatial Relationships: [YU97 ch. 11] deducing spatial relationships
- Texture: [MA96] Match based on texture image patterns and Gabor wavelets:
identify salient locations (edges, corners, gray level changes),
compute local feature reps (56, from bank of Gabor filters),
cluster and represent by centroid feature vector
- Fractals: [ZHANA96] joint fractal coding between images,
using set of iconic images, clustering, R-tree-like indexing, sim
- Facial images: [WU94] CAFIIR
- 6 facial aspects (chin, hair, eyes, eyebrows, nose, mouth)
- principal component analysis coefficients vs. facial landmarks
- Self-Organizing Map, Learning Based on Experiences + Perspectives
- iconic index tree
- sim (correlation, distance, normalized correlation, combined)
- Image Retrieval
- Indexing: [BABU95] Color indexing: RGB or YUV, color clustering (using
histogram), color indexing (R-trees), test with trademarks
- Query Processing: [YU97]
- Combining Feature Contributions
- Browsing
- [WU96] trademarks: shape, multilingual words, fuzzy thesaurus, fusion
- [REMI97] with: nona-tree (block-oriented decomposition),
wavelets (Haar, Daubechies)
- Video Analysis
- Modeling
- Cues: motion (objects, camera),
frame-frame differences (sequence, offset)
- Hierarchy: scene, shot
- Stratification: overlapping, independent
- Construct: frame, attributes
- [ZHAN93] Automatic partitioning
- Difference metrics: pair-wise, likelihood ratio, histogram
- Graduate transitions: accumulated comparison, optical flow,
block matching
- Application: thresholds, multi-pass
- [SRIH94] CEDAR
- Capture and analysis: document understanding, page layout, OCR,
table interpretation, graph/chart recognition
- Topic categorization, photos plus captions (e.g., faces)
- Linkages: structural, temporal, conceptual
- [BULT95] CMIF - embedding video with other data types
- Versus: video on demand, video multicasting, video-based CSCW
- Applications: electronic encyclopedia, email, cookbook
- Embedding issues: constraints (presentation, sync, interaction),
scheduling and control (doc activation, channel support
- Representation: hierarchy of media events plus traversals
(parallel, sequential, hyperlink)
- [CHUA95] VRSS - Video Retrieval and Sequencing System
- Knowledge: thesaurus, cinematic KB
- Frame: start/end frame, type of shot, type of angle, actor list, prop list
- Scene editor and interface: list of scenes, scene hierarchy, shot list
- Shot editor and interface: shot list, icon area, video control
- Rules: parallel, concentration, enlarge, general, rhythm,
sequential, content
- [DIMI95] Motion Recovery for Video Content Classification
- MPEG macroblock trajectory through object motion tracking
- Hierarchies: spatial, temporal
- Hierarchy levels: semantic, semantic association, image features,
object descriptors, physical image
- Query language: EVA, plus visual interface VEVA
- [HAMP95] Production Model Based DV Segmentation
- Video production model: start with shot set and edit effect list;
assemble so have final cuts
- Edit effects as 2D image transformations: identity (cut), spatial
(page turn), chromatic (fades, dissolves), spatio-chromatic (wipe)
- Segmentation steps: feature extraction, classification, segmentation
- Feature detection: level 1 (histogram, difference image,
gradient image), level 2 (cuts: chi square histogram difference,
chromatic/spatial: image division, image constancy)
- Tuning: feature thresholds, error measures
- [KELL95] XMovie
- Architecture: CM server, CM client,
CM agent (in X server), playback app
- MTP: Motion Transmission Protocol
- Issues: DeltaCLUT, adaptive forward error correction
- [ZHAN95b] Compressed Data
- Using DCT coefficients: frame pairs, block comparison
- Motion vectors: moving objects, camera tilt, zoom
- Hybrid: together is better for gradual transitions, camera operations
- [LEE97] VIMS
- Video classification: key frame selection by net comparison (vs.
pairwise, likelihood, global histogram, local histogram) plus
seek and spread using wavelet coefficients
- Conceptual clustering: use role tree
- Hybrid: together is better for gradual transitions, camera operations
- Video-on-Demand
- Networkinig
- Disk layout, scheduling, use of tertiary storage
- Reservation schemes