Code
IGFE IMA 5001
Level
M2
Graduate
Graduate
Semester
Fall
Domain
Image
Program
Programme Ingénieur
Language
Anglais/English
ECTS Credits
4
Class hours
45
Workload
90
Program Manager(s)
Educational team
Organisation
Cours/TD/TP/projet/examen : 36/0/9/0Learning objectives
At the end of this module the students should be able to:
- Exploit the content-based representations with visual descriptors for indexing, searching and enriching heterogeneous content,
- Select appropriate descriptors for solving a given retrieval application, and integrate it within a content-based search engine,
- Analyze and evaluate interoperable indexing tools based on multimedia standards and description languages (MPEG-7, XML…),
- Conceive and implement a semantic learning chain for a given semantic indexing application, focalized on a specific type of content, including feature extraction, descriptor specification and machine learning stages,
- Construct a deep learning approach for solving an artificial intelligence-related problem (e.g. image/object recognition).
CDIO Skills
- 1.1.1 - Mathematics (including statistics)
- 1.2 - Core engineering fundamental knowledge and other disciplines
- 1.3 - Advanced engineering fundamental knowledge, methods and tools
- 2.1.4 - Analysis With Uncertainty
- 2.1.5 - Solution and Recommendation
Prerequisites
None
Keywords
Multimedia indexation, visual descriptors, shape, color, motion, texture, description scheme, description languages, MPEG-7standard, semantic learning, object recognition, deep learning
Content
- 2D/3D shape extraction
- Color representions
- Extraction of texture primitives
- Motion analysis
- The metadata era: a new multimedia consumption
- Low-level descriptors for content indexing and content-based access: color descriptors, shape descriptors (2D, 3D and 2D/3D), motion descriptors, texture descriptors
- Overview of the MPEG-7 standard
- Interest point descriptors
- Automatic extraction of semantics: from pixels to semantic entities based on machine learning approaches
- Artificial intelligence: deep learning techniques
- Video structuring and abstraction: shot/scene detection, keyframe selection
- Object-based representations
- Query by example and similarity metrics
- Towards high-level descriptions: description schemes, hierarchical and multigranular representations
- Structural and semantic descriptions of multimedia documents
- Training, profiles and relevance feedback
- Search engines and data mining
- Deep learning
- Applications (video archiving, sign language, face recognition)
Evaluation
Continuous evaluation based on lab assignments (BE) and personal supervised project (P) linked to real industrial applications or to national/European research projects.
Final mark = Average (BE, P)