• Årsstudier
• Bachelorstudier
• Master Programmes
  • Information Security
  • Media Technology
  • Master's Thesis
  • Courses
• Student exchange
• Om IMT
• Research

Media Technology 2007

Student Name	Thesis Title	Assigned By	Teaching Supervisor
Berg, Amund	Bruk av eyetracking for evaluering av visuelt komplekse TV-programmer	Mediatechlab	Frode Volden
Carson, Roger	Comparing item-based collaborative and demographic group recommending methods	GUC	Rune Hjelsvold
Dugay, Fabienne	Evaluation of gamut mapping algorithms	Colorlab	Jon Yngve Hardeberg, Ivar Farup
Gustavsen, Eirik Reier	Classifying motion picture audio	Mediatechlab	Faouzi Alaya-Cheikh
Halvorsen, Martin	Content based lecture videos indexing	Mediatechlab	Faouzi Alaya-Cheikh/Rune Hjelsvold
Jansen, Yorrick	Video analysis (TBD)	Colorlab	Faouzi Alaya-Cheikh
Kominkova, Barbora	Perceptual image difference	Colorlab	Eriko Bando/Jon Yngve Hardeberg
Lefloch, Damien	Video analysis (TBD)	Colorlab	Faouzi Alaya-Cheikh
Mauduit, Nicolas	People counting using video processing	Colorlab	Faouzi Alaya-Cheikh/AI
Mikalsen, Espen Bårdsnes	Verification and extention of a camera based calibration method for projector displays	Colorlab	Jon Yngve Hardeberg/Jean-Baptiste Thomas
Narvesen, Håvard Gjære	Personalisering av læringsmateriale i nettbasert opplæring	Mediatechlab	Rune Hjelsvold
Nielsen, Mogens	Functionality in a second generation tag cload	Mediatechlab	Rune Hjelsvold
Panák, Ondrej	Color matching under soft-proofing conditions	Colorlab	Peter Nussbaum/Jon Yngve Hardeberg External supervisor prof. RNDr. Marie Kaplanová, CSc.
Paul, Steffen	Color management in digital intermediate movie production	Colorlab	Arne Magnus Bakke/Jon Yngve Hardeberg
Pedersen, Marius	Importance of region-of-interest on image difference metrics	Colorlab	Jon Yngve Hardeberg, Peter Nussbaum
Reinsby, Mats	Using handheld devices as a tool for learning	GUC	Rune Hjelsvold
Renani, Siavash Asgari	Characterisation of an imperfect screen	NEC Corporation, Japan	Jon Yngve Hardeberg
Sagbakken, Hans Christian	Irisgjenkjenning under varierende forhold	GUC	Faouzi Alaya-Cheikh
Savalle, Mathias	People counting using video processing	Colorlab	Faouzi Alaya-Cheikh/AI
Øvern, Jon Anders	Film restoration using ACE extensions	Colorlab	Ivar Farup/Jon Yngve Hardeberg

Abstract/Sammendrag

Berg, Amund

Bruk av eyetracking for evaluering av visuelt komplekse TV-programmer

• Bruk av eyetracking for evaluering av visuelt komplekse TV-programmer (ppt)
• Thesis
• Attachment Attachment (zip)

Det finnes lite kunnskap om hvordan TV-seeren ser på skjermen og hvordan dynamisk grafikk i bildet påvirker seerens evne til å oppfatte hovedbudskapet. Gjennom dette prosjektet vil eyetracking bli brukt for å finne hvor seeren fester blikket når ulike former for dynamisk grafikk blir lagt til på en TVproduksjon.

Måten TV-programmer produseres er under utvikling. Med større behov for og bedre muligheter til å legge dynamisk grafikk i bildet er det interessant å vite hvor mye og hvilken av denne grafikken den vanlige TV-seeren får med seg. Hovedformålet for denne oppgaven er å finne hvor store mengder dynamisk grafisk innhold som kan benyttes i en TV-produksjon før seeren mister oversikten og slutte å få med seg budskapet. Eyetracking utstyr vil bli brukt for å gjøre eksperimenter.

TV-programmer blir tradisjonelt sett produsert med en strøm av bilder med noe statisk grafikk som overlegg. Det blir stadig mer vanlig at TV-programmene blir mer visuelt komplekse og gjerne begynner å nærme seg det vi har sett i brukergrensesnitt på web. Skjermen blir delt opp i flere rammer med dynamisk innhold. Vi ser dette for eksempel på SMS-TV hvor hovedvideostrømmen er plassert i en ramme, en chat hvor seeren kan sende inn meldinger i en annen ramme og flere rammer med forskjellig rullende tekst blir plassert rundt disse. Andre eksempler er sportssendinger hvor resultater, statistikk og annen informasjon vises, eller politiske debatter hvor seernes kommentarer og respons vises. Det er liten kunnskap om hva disse visuelt mer avanserte formatene har å si for seeropplevelsen og effektiviteten for produsent – seer kommunikasjonen.

Eyetracking har blitt et godt verktøy for forskere som ønsker å finne ut hvordan seeren bruker oppmerksomhet brukes for å tolke brukergrensesnitt, tekstdokumenter eller bilder. Eyetrackingutstyr kan også brukes som et verktøy for å få ny innsikt i hvordan seeren oppfatter visuelt komplekse TV-programmer. Dette prosjektet vil være en del av TiDE-prosjektet (Television in Digital Enviroment) ved Høgskolen i Lillehammer.

Tid er penger i TV, spesielt for reklamebransjen. Derfor er det viktig at informasjonen som vises faktisk blir oppfattet av seeren. Når en produsent velger å legge informasjon inn i TV-bildet gjøres det for at seeren skal få det med seg. Det er derfor interessant å vite hvor mye seeren greier å få med seg av det som blir lagt til og hvor stor nytte seeren føler han har av informasjonen. En annen motivasjonsfaktor for prosjektet er muligheten for å vurdere eyetrackingutstyret i praktisk bruk i seerundersøkelser for TV. Dersom nytteverdien av eyetrakingen overstiger den påvirkningen det tekniske utstyret har på deltakeren i prosjektet vil det ha betydning for hvordan seerundersøkelser kan gjennomføres.

Forskningsspørsmål

1. Er det en sammenheng mellom forflytting av fokus og seeropplevelse?
   Ut fra eyetrackingen får man resultater som beskriver hvor seeren har sett til forskjellig tid. Om man tar utgangspunkt i disse resultatene og sammenligner med resultater fra andre former for undersøkelser, slik som spørreskjema eller intervju, kan det da påvises en sammenheng mellom hvor mye seeren flytter blikket rundt på skjermen og hvor fornøyd seeren er med opplevelsen?
2. Hvordan påvirkes seeropplevelsen av dynamisk grafisk innhold?
   Dersom man tar utgangspunkt i et rent TV-bilde og legger til forskjellige typer og mengde av dynamisk grafisk innhold på en slik måte at resultatet blir mer og mer visuelt kompekst, hvordan vil dette påvirke seeropplevelsen? Vil dette ha noen virkning på i hvilken grad seeren får med seg hovedbudskapet i et TV-program? Er det noen markante forskjeller for eksempel i alder eller kjønn? Hvor mye og hvilken type dynamisk innhold kan legges til uten at seerens oppmerksomhet blir fjernet fra hovedinnholdet?
3. Hvor godt vil eyetracking-utstyr være for å måle seeropplevelse?
   Hvor godt eyetracking-utstyret fungerer for å måle seeropplevelse vil være avhengig av blant annet hvor transparent teknologien vil være for seeren. Er teknologien for begrensende i forhold til hvor seeren kan se og bevege seg kan dette være noe som gjør at seeren føler det er en urealistisk situasjon noe som vil påvirke resultatene. Godt vill i denne sammenheng si at det er en god sammenheng mellom de resultater man får ut av eyetrackingen og de resultatene andre former for måling av seeropplevelse gir.

Carson, Roger

Comparing item-based collaborative and demographic group recommending methods

• Personalized information in a group recommender system (ppt)
• Thesis (pdf)

Consumers are every day surrounded by pamphlets, brochures that are sent in the post. There are large posters everywhere, adverts on tv, radio, internet. Companies spend a lot of money on marketing, and this is where Marketing Communication comes inn. Marketing communication comprehends allmethods of communication between a company andthebuyers[1].Personalized marketingisnotanew phenomenon, companieshave for many years sent out flyers and brochures to customers with their names on them. AllSkilt is going to design a service called MentonIT that will amongst other things give personalized advert’s to the users. This personalization is going to be done by using the users own personal profiles to choose the best advert for the user. The advert will be displayed on a display of some kind and these displays will in many cases be located in a public setting. This means that there in many cases will be more then one user in front of the screen at any given time. Group recommendation then becomes an issue. The recommended advertswillhaveto chosenonthe basisofthe personalprofilesofall the users in front of the display.

Since this topic is a little explored one and the need for a stable and reliable solution to the problem is an ever growing one. There has been done extensive research in the field of the recommendations for individuals, but very little on the recommendations for a group. The research done in this thesis will not only be usable for the marketing side of recommendations.Itcaneasilybeadaptedtogive personalized informationtoagroupof people waitingat the train or buss station. It could be used for any public setting where it is desirable to present information targeted at an entire group of people.

This is a difficult topic, but an interesting one. This master thesis has the possibility of being part of a unique and groundbreaking service for the consumers of the world. The topic of group recommendations has only been explored to a certain extent. Group recommendations based on the users personal profiles is even less explored. Most of the recommender systems focus on the creating recommendations for individuals. In addition the earlier mentioned obstacles, this particular problem also means the users have no direct way of giving the system feedback on the recommendations they are given. According to AllSkilt there are no other similar services or systems available at this time. This service can potentially redefine the way marketing communication is done. Consumers, companies and producers throughout the world could in the future be affected by the research done in this thesis and on through the development.

By exploring the topic of group recommendations based on the personal profiles of the individuals the thesis will help to create a better selection of adverts or information for a selected group of people. The consumers, the users of the service will want to get as personalized adverts as possible. The companies and producers want to give the right adverts to the right people, in thehope that they will buy their product/service.

Research question: Find a way to effectively present relevant information to a group based, where the information was suitable for the individuals in the group.

Dugay, Fabienne

Evaluation of gamut mapping algorithms

Fullu, Christopher Johnsrud

Highlights extraction from unscripted video

The process of extracting highlights from an unscripted video source is more complicated than highlight extraction from a scripted video source. The unscripted source could be a sports video, e.g. a football game, whereas the scripted content could be e.g. a movie. In a football game the video shots are very similar to each other, which results in unreliable shot detection. However the position on the field as well as trying to detect if many players are close to each other may indicate a highlight. The audio track or commentary of a field sport may be of great help. The analysis of the audio track may help reducing the number a false markers, and help segment the sports video. For example, when a player scores a goal in football, the referee blows his whistle, the crowd watching the game may make a lot of noise and the commentator usually screams higher than usual. This thesis will take advantage of both audio and visual details in any unscripted video source, e.g. sports, video, and develop a framework that can analyze these sources and extract highlights. These highlights will be used to browse the unscripted video content. This thesis will concentrate on sports videos from field sports.

Trying to partition/segment a video from an unscripted source is more complicated than from a scripted source. This is mainly due to the unreliability of shot detection in unscripted content. For example in a football video the visual features are often very similar in most of the frames, which makes grouping them into different shots a difficult task even when there are clear semantic units within the same period such as attacks on the goal, and counter attacks in the mid-field. In the earlier years of video retrieval by content the audio track of a video was often ignored. The audio content of an unscripted video such as a football video often contains valuable information that can be used to achieve better temporal segmentation of the video. For example in a football game, any highlights are usually notable in the audio. If a team scores a goal, the crowd watching the game may start making some more noise than usual, the referee may blow his whistle, and the commentary track may become more intense. This can be used to reduce the number of false alarms when segmenting an unscripted video, and therefore help performance of such a system. Such a system should contain:

• Video analyze system, that can determine the position on a field, e.g. midfield
• or goal area, and if there are many players packed together, witch may indicate a highlight
• Audio analyze system, which can differentiate such things as the intensity of the commentators voice or a referees whistle
• Combination of these two systems to reduce false detections and therefore maybe increase performance
• Algorithms for highlight extraction, based on the markers created by video and audio segmentation.

There are several articles on extracting highlights from sports video, not so many however deals with both audio and visual features at the same time. Most projects or articles focuses on either one of them, as apposed to earlier times in video segmentation where only visual features were considered, some does now look at the audio features of the unscripted video. Some projects have been done witch combines both techniques, so a contribution and further research on how to further improve these algorithms and techniques would be valuable.

The goal of this thesis should be to contribute to the ongoing development of audio-visual segmentation, and create a framework that can extract highlights from unscripted video. Propose new algorithms to increase performance and reduce the number of false detections or missed detection should be the focus, after a general framework build upon existing work has been implemented. Problems as distinguishing goals from missed shot-on-goal and general effects from camera motions would be problems that should be worked on, and analysis of the excitement levels in audio as well as tracking of specific sound like a referees whistle.

Research questions

1. How one best should utilize the existing techniques to achieve as few false detection or missed detections as possible?
2. How to use cues from the audio and the visual tracks to achieve better highlight detection?
   Based on the limitations of exiting algorithms, what is the best way to go when trying to design algorithms to better extract or more exactly extract audio and video markers?
3. How to link low-level audio-visual features to high-level events in the video sequence (semantic gap bridging)?
   Based on the limitations of highlight extraction, such as a missed on-goal-shot in a football match being detected as a highlight, what is the best way to enhance or develop new algorithms for highlight extraction?

Gustavsen, Eirik Reier

Classifying motion picture audio

• Classifying motion picture audio (ppt)
• Thesis (pdf)

These days there are an enormous amount video material available to everyone. The challenge for most people is to find out anything about the content just by reading the tag lines, plots, ratings and reviews. The plot of a movie tells something about the story, not how the story is presented. The ratings are often based on the presence of different features. And reviews are based on one individual’s opinion of the movie. It is seldom that you can find out how much violence, how much dialog, how much music and how much silence there is in the movie. There is also little information about what kind of music, what kind of dialog and if it’s a noisy or a quiet sound image. Much of this could be helpful to know before deciding to watch a movie or not. Or perhaps, decide if it is suitable for a child to watch.

Content analyses, and categorizations, based entirely on video have sometimes failed just because of that they are only based on the video content. Dialogs are not detected because the people who speak are not visible, nudity filters have blocked documentaries about Indians in the rainforest and humour has been mistaken for violence. In these instances, audio analysis could have contributed to categorize the content differentially.

It is possible to extract a lot of different features form the audio track, but would it be possible to make a meaningful table of content? And would it be possible to compare the different tables? The question to be answered is: Is it possible to combine algorithms for extracting different features in such a way that it would be possible to say something about the semantic content in a movie?

Problem/Question

1. How best to classify audio segments into for instance: Silence, Music, Speech and Noise.
2. How to detect dialogs and dialog tone based on the audio contents
3. Based on the extracted audio features and the segment classification, how to best classify the video content into the different genres?
4. Create a table of contents of the video, based on the audio track only, to describe the video content.

Halvorsen, Martin

Content based lecture videos indexing

• Content based lecture videos indexing (ppt)
• Thesis (pdf)

All around the world there are students who live in another geographical area than their educational establishment where their lectures are held. Both students that have a long travel route and distance-learning students may need other ways to be educated. E-learning systems are an approach to educate students that are not able to attend in lectures. By using multimedia content students are able to be educated anywhere.

Many educational establishments make lecture notes and other related syllabus on the Internet. In most cases PowerPoint slides and other notes make up the basis for the educational material. However, the information given in a lecture is a lot more in-depth than the information published, and surveys shows that students gain more in attending a lecture than only reading published lecture notes.

Many establishments want to offer more, such as multimedia content, to each student. Multimedia content makes the material accessible at all times for students that for instance need a revision or for those who missed a lecture. Videos of lectures could be an important source to educational information, especially in distance teaching. Even though both the expertise and equipment to create a summary video of a lecture is available, it might be time consuming and not prioritised by some teachers. Without any index, the video may be hard to navigate in. In for instance a process of revision, you have to see the whole video and manually navigate in order to find the specific information you want. This is not a user-friendly option and for some it will be easy to give up. Therefore, a simple video that is easy to navigate in may need a lot of pre-processing before it can be published.

With the help of automatic systems it is possible to create a navigable lecture video without manual editing. This will be a useful tool for all educational establishments that wish to publish lecture videos for their students or course participants in an effective and easy way.

A large amount of educational establishments don’t have required equipment or the expertise to create a user-friendly lecture video. Many videos that are created have no index and are hard to navigate in. Some requirements have to be fulfilled in order to make students use the lecture videos, and have utility value from them; the video has to have good light conditions, good sound and is navigable, with for instance an index. By the use of technology, it is possible to solve the indexing problem. However, technology may not improve individual skills concerning other aspects of movie making.

There is a reason to believe that educational establishments that offer lecture videos may have a benefit in attracting more students, and different types of students. For instance, this offer may be highly relevant for distance learning students and student that for different reasons can’t or won’t attend ordinary lectures. On a general basis, all students may benefit from this by having all lectures accessible at all times for revision purposes.

An automatic process for creating an index for a lecture video should be of interest world wide. If the process creates an index that’s valuable for students in finding or revising information, more establishments should be able to create good lecture video summaries with little effort. By contributing in this field, others may use it as building blocks to create better and more complete systems.

Research questions

Q1: How can foreground/background segmentation in a lecture video work for different writing-boards? When taking different writing-boards into account, colours and sizes of the board cannot be used. There could also be other people entering the scene. The main use for this segmentation is to segment a teacher in front of a writing-board. There should be as few static variables as possible to make the algorithm general for lecture videos with textual content.

Q2: How to automatically extract meta-data from lecture videos? The definition of meta-data in this contest is “what is happening”. In a lecture video several things can be included as a meta-data; when teacher’s writing (increase of ink/chalk), cleaning board (wipe out blackboard, decrease of chalk, turn page on a flip-over) and when teacher is talking/asking/discussing (no increase or decrease of ink/chalk over a time period). Other meta-data may be figures and headlines. Meta-data can be useful to use as indexes in a lecture video.

Q3: How to use such meta-data for indexing and searching of lecture videos? A good navigation in a lecture video is important to find information one look for. By using meta-data in indexing there would be possible to make it easier to find information. Different approaches will be discussed on how to index, or maybe create textual meta-data or description of some of the meta-data extracted. The textual meta-data can furthermore be used in textual searching.

Jansen, Yorrick

Video analysis (TBD)

Jbilat, Rhanem

Robust Estimation of Colour Gamut Boundaries in the Presence of Noise

Kominkova, Barbora

Perceptual image difference

Lefloch, Damien

Video analysis (TBD)

Mauduit, Nicolas

People counting using video processing

Mikalsen, Espen Bårdsnes

Verification and extention of a camera based calibration method for projector displays

• Projector calibration with colorimetric camera (ppt)
• Thesis (pdf)

The prevalent method of doing presentations today is electronic presentations. The use of projection display technology has eased the task of showing a crowd a complex output with contents that go far beyond text materials. Figures, illustrations, images and videos are all graphical output that depends on the display devices color reproduction. Far too often these kinds of illustrations are reproduced incorrectly due to lack of device calibration. This can be critical when the graphical contents purpose is to illustrate an important message. Examples can be business and marketing presentations, educational presentations or technical presentations. The effect of mismatch in output can be a sever distortion of the actual presented text and illustrations. This problem occurs due to a non-consistency in color reproduction. Colors reproduction is device dependant. This means that different color imaging devices reproduce colors differently. This unwanted change in visual appearance is often exemplified by how something can look good on a computer display, but not when reproduced with a printer. But the effect is also true between different display devices and even between displays of the same type. This is often a result of each display devices characteristics. A permanent projection setup such as a home theater, or a dedicated presentationroom are often tweaked to produce best results possible in stable conditions. This is usually not the case for the more portable projectors systems. These are usually smaller units that may frequently change location, and thereby a change in display conditions, like luminance in room and surface projected on. The problem of non-consistency apply to all use of projectors. But may be more apparent when a projector is often set up in more occasional and changing environments. In these cases adjustments to achieve better color reproduction are rarely done. This results in a what-you-see-is-what-you-get result. It has been proven that a camera can be used as a radiometric measurements device[1, 2] to replace the function of more expensive measurement equipment e.g. a spectrophotometer. If the aim is to implement a device for measurement with a projector. Then a digital camera could be a cheap, but effective replacement.

Problem description

The problem handled in this thesis is to improve users capability to adjust color reproduction and settings when using a projector. An aim in using such techniques is to enhance image appearance with the use of technology instead of traditional visual adjustments. Here the problem is approached through using a digital color cameras vision to evaluate projected image.

In such a setup the camera and attached system should be able to evaluate projected lightness, contrast, and the chromatic reproduction of colors. With basis in cameras observations, calculations should be made for adjustment of projection units output and accounted for.

Narvesen, Håvard Gjære

Personalisering av læringsmateriale i nettbasert opplæring

• Personalisering av læringsinnhold i LMS (ppt)
• Thesis (pdf)

Behovet for hva man bør igjennom i et kurs, avhenger ofte av spesialisering og individuell kunnskap. Ulike grupper kan ha ulike utgangspunkt og læringsprofiler. Kursdeltakeren må ikke motta for mye kjent informasjon, ei heller motta informasjon som forutsetter kunnskap som kursdeltakeren ikke har. Læringen bør tilpasses den enkelte kursdeltaker. Utfordringen i slike tilfeller er å bestemme hvordan kursinformasjonen skal presenteres for de ulike individene. Masteroppgaven skal starte med et litteraturstudium innen brukertilpasning i «Adaptive Learning Management Systems». Dette studiet danner fundamentet for designet av en prototyp på et kurssystem. Kurssystemet skal være i stand til å personalisere læringsinnholdet i et kurs basert på kursdeltakerens forhåndskunnskap om emnet. Med prototypen som verktøy, skal det gjennomføres et eksperiment med to brukergrupper. Den ene gruppen skal gjennomføre et enkelt kurs med tilpasset læringsinnhold, mens den andre gruppen tilbys en «one size fits all»-tilnærming av det samme kurset. Eksperimentet gjennomføres for å si noe om nytteverdien av brukertilpasset servering av læringsobjekter. Nøkkelord: e-læring, personalisering, «Learning Management Systems»

Mennesker er ulike av natur og lærer på forskjellige måter. Forhåndskunnskap, interesser og ferdigheter er parametere som påvirker behovet for læring. Læring er en personalisert og adaptiv prosess som fra start til mål burde ivareta den individuelle brukers behov. Målet med alle kurs er at kursdeltakere skal være på et bestemt kunnskapsnivå når kurset avsluttes. Ofte kontrolleres dette ved hjelp av en test eller eksamen. I det vanlige klasserommet vil en lærer forsøke å gi elevene oppgaver som er hensiktsmessig for å komme opp på det nivået som kreves. Hvordan en lærer bestemmer seg for hensiktsmessige metoder, avhenger ofte av kunnskap om elevene. Hvor mye kan de fra før? Hvordan er progresjonen? På tilsvarende måte bør også et e-læringsprogram kartlegge læringsprofilen til den enkelte kursdeltaker.

Forskningsspørsmål

1. Hvordan bør et kurs tilrettelegges slik at kursdeltakeren får tilpasset læringsobjekter fra en pre­test? Med tilrettelegging menes å foreslå læringsmateriell i form av «Learning Objects» (LO) som passer kursdeltakerens kompetansegap. Et kompetansegap kan beskrives som det påkrevde kunnskapsnivå for å bestå det aktuelle kurset minus den kunnskap kursdeltakeren allerede besitter. Pre-testen skal brukes til å avdekke kompetansegapet til den enkelte kursdeltaker. Studien skal ende med utvikling av en prototyp for gjennomføring av et brukertilpasset kurs.
2. Neste steg består i å undersøke brukertilpasset servering av læringsobjekter i kurs kontra ikke­tilpasset kurs. Med prototypen som verktøy er det ønskelig å finne ut om det er noen forskjell på: o kursresultat (eksamenskarakter) målt i forhold til tidsbruk mellom et tilpasset og ikke­tilpasset kurs? o hvor fornøyde kursdeltakerne er ved et tilpasset og et ikke-tilpasset kurs?

Nielsen, Mogens

Functionality in a second generation tag cload

• Improvements of tag-clouds - A survey (ppt)
• Thesis (pdf)

I forbindelse med den økende populariteten av tags (tilegning av nøkkelord til bilder, lyd, nettsider etc.) har det blitt brukt "tag-clouds" (heretter kalt tagskyer) for å visualisere taggene forbundet med informasjonen. Disse tagskyene er ikke alltid tilfredstillende for informasjonsgjenfinning og de har blitt både skrytt av og kritisert. For å forbedre tagskyene har det blitt foreslått flere møter å endre på måten de viser informasjon og hvordan en bruker kan interagere med dem. Sider som Flickr og del.icio.us har vist at tagging er veldig nyttig som en hurtig, enkel og brukervennlig måte å beskrive bilder og nettsider. Ved å tagge sine digitale ressurser (bilder, nettsider eller lignende), kan en bruker enkelt finne igjen disse fra en hvilken som internettilkoblet datamaskin. I tillegg har nettsteder som for eksempel Flickr og del.icio.us også tilføyet et sosialt aspekt til taggingsprosessen - på disse nettstedene kan forskjellige brukere se andre brukeres lagrede ressurser ved å søke i taggene som har blitt brukt av de mange brukerne. Dette sosiale nettverket gjør det også mulig for brukerne å utnytte andres tagger til å oppdage nye og interessante ressurser på Internettet.

Selve taggeprosessen er oftest ukontrollert og brukeren kan selv velge sine tags blant hele sitt tilgjengelige ordforråd, dette gjør det til en enkel oppgave å tagge en ressurs. Dette sammen med at taggingen gjør gjenfinningen enkelere for brukeren og at brukeren ser en sosial nytte i å tagge gir en lav inngangsterskel for nye taggere.

Med en stor mengde brukere og data tilgjengelig på slike nettsteder, er det interessant å kunne visualisere all denne informasjonen på en måte som kan gi nye brukere en hurtig og intuitiv oversikt over hva som er tilgjengelig og ikke minst hvilke tagger nettstedets brukere har valgt mest. Denne visualiseringen gjøres mulig ved hjelp av tagskyer. En tagsky er en liste over de mest populære taggene, vanligvis fremvist i alfabetisk rekkefølge og visuelt vektet ved hjelp av skriftstørrelse (fontstørrelsen øker i takt med taggens popularitet). Dette gir en mengde av ord som er brukt som tagger hvor de mest brukte bobler til overflaten. Når en bruker så klikker på en tag vil han bli vist de ressursene som har blitt tagget med dette ordet.

Etterhvert som tagskyer har blitt mer og mer populært som visualiseringsverktøy, har det også blitt kritisert for sin visualiseringsmåte og det har blitt foreslått flere måter å forbedre dem på. Denne rapporten vil presentere forskjellige forslag til informasjonsvisualisering i forbindelse med taggesystemer og styrker og svakheter med de forskjellige forbedringsstragegiene. Ved å presentere og analysere disse forbedringene, vil man kunne skape et orden i kaoset og gjøre det lettere for utviklere å finne ut hvordan de kan visulisere sin informasjon til sine brukere.

Panák, Ondrej

Colour memory match under disparate viewing condition

In the previous works were found significant trends in memorising of a different colours, and trends in the colour memory shift in a long-term and a short-term colour memory were described. There were also investigated image context factors and time factors in colour memory experiments. Difference between instant memory colour and later memory colour is called colour memory shift. This memory shift could be explain by cognitive effect, and in literature is mentioned that is not related to the viewing condition. Some works were focused on investigation of colour memory, by selection colours under two different light sources (A,D65) and dissimilarities were found. The distinction between perception and cognition is not clear.

Aim of this project will be to find, if the changes in variables in the colour appearance model (CIECAM02) have some noticeable influences on the colour memory match and if any trend can be seen. It is expected, if the colour memory is not related to the viewing condition, there should be not significant differences in CAM space coordinates for different viewing condition. To find the answers, the psycho-physical colour matching experiments will be made.

Paul, Steffen

Color management in digital intermediate movie production

Pedersen, Marius

Importance of region-of-interest on image difference metrics

• The importance of regions in color image difference metrics (ppt)
• Thesis (pdf)

Today there is no good way of predicting perceived image quality without doing a psychophysical experiment. This is both time consuming and resource demanding. The color image difference metrics available today have been proved shortcoming when it comes to predict perceived image quality. When an observers view an image some regions are more important than others, some regions are important to some people and other regions to other people. The regions that are important and draws most attention should be weighted higher in a pixel-by-pixel image difference metric.

If important regions could be identified, changes in an image could be done according to the importance of the regions, this could then improve the overall perceived image quality.

The aim of this project is to investigate the importance of regions on color image difference metrics. A comparison of different regions from the eye- tracker, the observer and predefined regions will also be carried out. Regions from individual observers will also be used to see if there is a correlation between these and the individual results from each observer.

Many people, both in the graphic art and the average user would like to predict image quality. This would eliminate the use of time consuming psychophysical experiments. When someone changes an image, how will this change be perceived by the user? Will it be just noticeable, be an artifact or not perceivable at all? There are many image difference metrics available today, but neither of these are able to predict perceived image quality very well. When a change is made in an image, which regions of the images does the observers pay attention to? Could a great color difference in some regions be acceptable and just noticeable, while in other regions small color difference will be well perceivable? Knowing this in beforehand can help us to predict perceived image quality.

Research questions

• Q1: Can regions improve color image difference metrics in complex images?
• Q2: Comparison of the different regions, does observer regions correlate with the gaze regions and changed parts of the image?
• Q3: Can individual region-of-interest be applied to image difference metrics to correlate with the observer’s perceived quality?
• Q4: Which criteria do the observers build their decision of image quality on?

Reinsby, Mats

Using handheld devices as a tool for learning

• Using handheld devices as a tool for learning.ppt (ppt)
• Thesis (pdf)

e-Learning has become a big industry, and several companies have based their entire business model on the development of systems for electronic learning. A few years ago these courses were most commonly distributed using CD-ROM, but today the e-Learning industry is highly dependent on the World Wide Web (www) as a publishing media for their content. Learning Management Systems (LMS) are usually the backbone of most of these systems. An LMS is a web based application that keeps track of active learners, their progress in each course they are attending, and the actual learning content. This content has up until recently been shaped explicitly for each LMS, and reuse of course material between different vendors was nearly impossible because of the lack of an industry standard.

In 1997 the Department of Defence (DoD) in the USA formed a group called Advanced Distributed Learning (ADL) Initiative. Their goal was to “Provide access to the highest quality education, training and performance aiding, tailored to individual needs, delivered cost effectively, anytime and anywhere”. As a step towards their ultimate goal ADL developed the Sharable Content Object Reference Model (SCORM) by combining several standards from standardization organizations like Institute of Electrical and Electronics Engineers (IEEE), IMS Global Learning Consortium, Inc. (IMS), Aviation Industry CBT Committee (AICC) and Alliance of Remote Instructional Authoring & Distribution Networks for Europe (ARIADNE). SCORM defines how to package, transmit and launch learning material within an LMS. By providing a standardized way to handle learning content for e-Learning, ADL hoped to achieve reusability of learning content between different LMSs, and to make their Sharable Content Objects (SCO) synonymous with the term Learning Object within the e-Learning industry. Hungry for a standard, several companies adopted the SCORM specifications, and ADL has become a leading star in the development of e-Learning and Learning Objects.

During the last few years, however, a lot of researchers and educators have begun to question ADLs role in this development. This is mainly because of their obvious connection to the American Army, but also because of the lack of competition from other standards. ADL officials have stated that “…SCORM is designed for single-learners…”, but e-Learning today is moving more and more towards collaborative learning using tools as online communities and mobile phones. There are more than 2.7 billion mobile phone users in the world today, which is more than three times the number of Personal Computer (PC) owners. Knowing this it’s hard to imagine a future where handheld devices won’t be used in most contexts where the PC has been used for years, including learning.

The question is if SCORM is the right way to go when developing systems that support the use of handheld devices for learning. The project will investigate different factors that are important to decide whether or not learning content is suitable for presentation on mobile devices, and try to relate this to the existing SCORM standard.

Renani, Asgari

Characterisation of an imperfect screen

How a color is reproduced depends on many things. The colors are device-dependent, environment-dependent and so on. Achieving a color constancy over different systems have been a research goal in color science. This has obvious advantages; The artist designing his art on a computer can show his art as he intended to be perceived on a projector or other medium.

Projectors have become more and more a product not only for the industries but also for the common man. The prices are getting lower and the quality is getting better. The advantages of a projector in comparison with other display devices are the ability to show a large image. The disadvantage is the price. When considering the price of the projector, one must also consider the cost of the screen which can (and often is) bought separate. The display quality is a product of both projector and screen. A low quality screen yields bad image. In the recent years, several displaying systems have been developed using a projector as basic building block. To align the images and to avoid shadow-effects cameras are used. Using cameras as input mechanism for the system also makes the system dependent on the camera and this makes the calibration and characterization of the camera an important issue.

When walls are used as a screen some of their texture shows up in images and the colour of a white paint on the wall is rarely conductive to give a good balance. Using an arbitrary surface as a projector screen means that one can not assume anything. There are no guarantees that the surface will have the colour of a preferred white colour or that it has a texture that will not interfere with the projected image. This inhibits the use of projectors greatly since the human perception is sensitive to this type of interference. To compensate for the textures the effect of the screen must be removed. The impact of the screen on the projected image can be captured by the camera. This requires a concise mapping between the image taken from the camera and the projected image. These mapping algorithms are often referred to as image registrar. These algorithms are dependent on the camera’s angel towards the projected image. The spot where the camera is located is referred to as the "sweet spot". Further steps are to localise the affected parts for each primary colour and increase or decrease the luminance based upon an average of the image. As the image registrar algorithms are dependent on the camera’s resolution and angle a query into this attributes will be done. One of the main questions is; how will the camera’s position affect the compensated image?

From previous work in this field it has been reported a problem with sharp edges. This is deemed to be because of the resolution of the camera. This thesis will also concern it self about how the resolution of the camera affects the results.

Sagbakken, Hans Christian

Iris recognition using mobile phone

• Presentation
• Thesis (pdf)

In information technology, biometric authentication refers to technologies that measure and analyze human physical and behavioural characteristics for authentication purposes. Examples of physical characteristics include fingerprints, eye retinas and irises, facial patterns and hand measurements, while examples of mostly behavioural characteristics include signature, gait and typing patterns. Voice is considered a mix of both physical and behavioural characteristics. However, it can be argued that all biometric traits share physical and behavioural aspects.

Biometric authentication is the technology of identifying a human based on his unique biological characteristics. Pin-codes, id-cards, passwords and credit-cards are keys to our identity. A common factor to all of them is the fact that they are easy to lose or to forget. Most of us find it hard to remember all the different password and pin-codes one needs in our modern everyday life. That's why its tempting to use the same access codes on the mobile phone, Internet bank and credit card. Which creates a high risk for all the systems used once the code is stolen. Biometric authentication provides us with codes that we cannot forget and don't need to remember.

The motivation to use of mobile phones for iris recognition is their accessibility and security. Nowadays, mobile phones are becoming versatile miniature personal computers, integrating all tools needed for capturing, processing and transferring complex media data. Therefore, they are being frequently used to perform complex tasks earlier possible only using fast computers equipped with special software and hardware. Image and video capturing, speech recognition, browsing the Internet, playing music, watching TV, GPS navigation are a few of such demanding tasks that mobile phones are used for.

Therefore, this master thesis focuses on using a mobile phone for irisbased identification. One possible usage scenario may be on-line banking systems. The banking system have a profile of the users stored, composed of full name, address, personal number, account number and the iris code. The mobile phone will be the interface between the users and the on-line banking system. It is also the navigating tool for browsing the bank's website. The user navigates to the account domain where he gets a message: "Write your personal number and submit". After he enters his personal number he is asked to: "Transmit your iris code". He than takes an image of his eye and the phone does all the processing for iris detection, iris code generated and transmission to the bank. The bank servers do iris code validation by comparing it against the client stored record in the customers database. The bank will allow the transaction or not based on the validity of the iris code match.

Research questions

Question 1: What minimal conditions are needed for this to be feasible? The first question focuses on the conditions in which the images are captured: i.e. lighting conditions, stability of the hands, non-frontal shots and out of plane shot.

Question 2: What are the minimal hardware requirements? In this part we want to focus on the minimal camera characteristics,processing power, memory and storage capabilities of the mobile phone required for the task.

Question 3: Which type of algorithms one can use for the preprocessing steps of the recognition? The algorithms need to be fast, simple and effective in memory usage.

Savalle, Mathias

People counting using video processing

Øvern, Jon Anders

Film restoration using extended ACE

• Film restoration using extended ACE (ppt)
• Thesis (pdf)

The state of early film recordings deteriorate rapidly, due to improper storage and the physical properties of the material used. This is regarded as very unfortunate, as much historical data will be lost if these recordings are not restored and archived properly, and soon. This is bad news for libraries and institutes that want to preserve these historical treasures. One of the problems is bleaching of the film, which makes colours less vibrant and changes the dynamic range of the frames. This process is chemically irreversible and can only be dealt with on a digital copy of the film, using film restoration tools. In 2005, Carlo Gatta presented his ACE (Automatic Color Equalizing) algorithm. ACE mimics the human visual system, using the concepts of colour/lightness constancy to equalize illumination intensity and colour cast in the image. This algorithm has several areas of use; one of them is film restoration. In his doctor thesis, he presents results from successful application of this algorithm on short film sequences. He states that the algorithm yields good results, but that it does not take temporal factors in consideration, and this can be improved. In general, the algorithm measures the difference between two pixels and uses a weight function to estimate how it influences the other. The comparison is done for every pixel to every pixel in the image. The second step is to map this into a new image. The effect is that it removes colour cast and improves lightness. Yet, some “unnatural elements” in films are intentional, but without the modifications for film restoration, the frames are balanced equally, changing illumination and removing colour cast. A film has many visual effects that this process will remove or disrupt, like intentional colour cast or fade out/in sequences. For more accurate results, an enhancement of ACE has been made. The enhancement carries out equalization, but has features that may be used to avoid disrupting the artistic properties of the film. But by solving a problem, you create a new one. With the new functionality one can restore images to their estimated intended appearance. However, frames that deviate a lot from their adjacent frames will no longer be equalized to the same illumination and colour, but will keep these properties after processing, as these deviations are interpreted as intended properties of the frame. This means that unwanted flickering will remain even after restoration. The challenge now is to find a way to do chromatic/spatial computation, keeping the visual properties intended, but also making the deviations of the damaged frames less significant. One way is to make ACE take inter-frame considerations. The first step of ACE is only meaningful to run on independent frames, but the dynamic mapping can be influenced by its previous and following frames. By averaging this mapping over a number of frames, deviations will become less significant, thus the flickering-effect will lessen. The goal is to further improve the algorithm to take care of more issues than it does in its current form. A functional implementation of the modified algorithm is planned, but the project will not include developing a fully-fledged automatic tool that can be put into service after project end.

Master's Thesis Media Technology 2007

© Høgskolen i Gjøvik,
Postboks 191, Teknologivn. 22, 2802 Gjøvik, Tlf. 61135100, Faks 61135170, E-post: postmottak@hig.no
Ved spørsmål eller kommentarer til websiden, webredaksjonen