Multimedia Information Retrieval
Transcription
Multimedia Information Retrieval
Multimedia Information Retrieval Toetsing Master AI/CS, MMIS, 2003, Lecture 1 Lecturer: Theo Gevers Lab: ISIS Email: [email protected] http://www.science.uva.nl/~gevers http://www.science.uva.nl/~gevers/master2003 De toetsing van het college is 1 schriftelijk tentamen en 1 opdracht. Het tentamen maakt 50% van het uiteindelijke cijfer uit. De opdracht de overige 50%. De opdracht is: Programmeeropdracht. Programmeeropdracht. Het implementeren van een beeldopzoek systeem. EN Schrijfopdracht. Schrijfopdracht. Het schrijven van ongeveer 10 pagina's over 1 bepaald onderwerp. 0 Preview Trends in Internet, Mobile Phones, Mobile Internet *source Nokia concept 0 Preview Trends in Internet, Mobile Phones, Mobile Internet *source Le Monde 300 Number of users in Europe in millions 250 200 150 100 50 1997 1998 1999 Mobile phone 0 Preview Trends in Internet, Mobile Phones, Mobile Internet *source Le Monde •There will be 200 million physical mobile users in Western Europe by 2005. 32 million of these users in Europe will be mobile multimedia users in 2005. The total Western European mobile market will be worth 104 billion ECU per year in 2005. •The mobile multimedia segment of this Western European market will be worth 24 billion ECU per year in 2005. 2000 2001 Internet 2002 2004 UMTS phone 0 Preview Trends in Internet, Mobile Phones, Mobile Internet Max speed GSM/WAP 9,6-14,4 kb/s Average speed Connection Stand by 9,6-10 kb/s circuit no GPRS 64 kb/s 20-50 kb/s packet yes EDGE 128 kb/s 80-110 kb/s packet yes UMTS 2 Mb/s 50-340 kb/s packet yes 1 0 Preview 0 Preview Trends in Internet, Mobile Phones, Mobile Internet 10Gb/s UMTS Speed 1Gb/s 1987 0 Preview Trends summary • • • • 1989 1992 SURFnet4 34 Mb/s 100kb/s SURFnet3 2M b/s 1Mb/s SURFnet2 64 kb/s 10Mb/s SURFnet1 9,6 kb/s 100Mb/s 1995 1998 2000 SURFnet5 20 Gb/s GPRS SURFnet5 2,5 Gb/s GSM/WAP SURFnet4 155 Mb/s Trends in Internet, Mobile Phones, Mobile Internet 2002 0 Preview Trends in senses of the Web SOUND The increase in the use of Internet The increase in size of digital photography (e.g. DV, Webcams) The increase of mobile services (UMTS phones) Digital TV 0 Preview Trends in senses of the Web SMELL •Shopping: stinking out loud •Gaming: the smell of victory •Entertainment: read and sniff •Television: getting to nose you •Media: Talking heads [beatnik.com] •Web browsing: Walls of sound •Advertising: Nowhere to hide •Health: Leveling the listening field 0 Preview What is the future? Multimedia information is the future 2 0 Preview Industry 0 Preview Multimedia information Visual worlds Company Application Data type Complexity Ilse Media Image retrieval on internet Visual/textual unstructured ANP Press Photostock Visual/category structured VNU/Nielsen Image tracking/monitoring Visual unstructured Akzo Nobel Coating effect classification Visual structured Forensic lab XTC-pill image classification Visual structured Philips People tracking Visual structured Elsevier Multimodal document retrieval Visual/textual unstructured SBS Billboard tracking/replacement Visual unstructured 0 Preview Demo1: real-time skin detection for human recognition 0 Preview Demo3: real-time object recognition and tracking *Hieu WWW Trademark assessment Trademark in public Video databases Product database general pictures 2D of 2D pictures 2D of 3D pictures general pictures limited domain unknown conditions known camera unknown conditions unknown camera known camera 0 Preview Demo2: skin/subtitle/speaker identification 0 Preview Demo4: real-time object recognition and tracking *Hieu 3 0 Preview Demo5: real-time human recognition and tracking *Hieu 0 Preview Demo6: real-time human recognition and tracking [Hieu, IEEE PAMI, 2003] Robust to background clutter and changing object appearance 0 Preview Demo7: real-time background detection and removal *Anuj 0 Preview Demo8: real-time object classification video material 0 Preview Demo9: real-time object classification video classification material shadow-shape classification shadow-shape 0 Preview Demo10: real-time object classification video classification 4 0 Preview Demo11: real-time object classification 0 Preview Demo13: real-time object classification 0 Preview Demo15: real-time object classification 0 Preview Demo12: real-time object classification 0 Preview Demo14: real-time object classification 0 Preview Demo16: real-time object classification 5 0 Preview 0 Preview Demo17: Mosaic Demo18: real-time object classification Feike Winkelman Techniques: • Mosaics. • Shot and key-frame detection. • Analysis of camera-motion. 0 Preview Demo19: real-time object classification: image serach engines 0 Preview Image segmentation: Philips Medical Systems [Ghebreab, IEEE PAMI] Interactive image segmentation to assist the user in outlining the objects in medical images • Techniques : Techniques: • Genre classification of image and video • Search and learning strategies in image and video databases • Interactive methods for image search 0 Preview Forensic search XTC-pills: Forensic Laboratory Interactive segmentation (medical images) 0 Preview Multimedia information Text-based image retrieval systems Convert content to keywords at insertion? run-time? once the query is given? Questions Key-words expressiveness Key-words ambiguity in language Key-words Digital image store 6 0 Preview Multimedia information Image database systems 0 Preview Example: content-based image retrieval query-by-example matching Query image Digital image Content-based image retrieval: query by example database query in natural format, unlimited repertoire Question computer vision is hard? how to handle in a database? 0 Preview Multimedia information Retrieval components Features text, colour, shape & texture Representation boolean, vector space and probabilistic Indexing inverted files, RS-trees Searching and finding k-nearest neighbor Interaction relevance feedback 7