87 research outputs found
License Management in Closed Offline Networks Using Modern Cryptographic Solutions
No full textNevion ønsket en løsning til et system som benytter seg av en tidsbasert lisensieringsmodell der kunden selv skal kunne generere del-lisenser av en større lisens på Nevions vegne, uten internettilkobling. Gjennom dette prosjektet har det blitt utviklet et konsept til det ønskede systemet ved å benytte seg av ulike teknologier som X.509, AES-256 og PKI. Videre har det også blitt implementert et bevis på at dette konseptet er mulig å gjennomføre i praksis. Håndtering av lisenser og digitale rettigheter i miljøer uten internettilkobling innebærer en stor sikkerhetsrisiko. Derfor ønsket Nevion å vite om et slikt system er sikkert, og etterspurte en sikkerhetsanalyse av det utviklede konseptet. Den totale løsningen gir Nevion et godt utgangspunkt til å implementere et liknende system og benytte prosjektrapporten til videre forskning og utvikling.Nevion wanted a concept solution to a system that incorporates a time-based licensing model, where the customer can generate sublicenses, from a larger top level license, on the behalf of Nevion. This should happen without an internet connection. Through this project, there has been developed a concept to the wanted solution through combining technologies like X.509, AES-256 and PKI. Furthermore, there have been developed and implemented a proof of concept, proving the practicality aspect of the system. Managing licenses and digital rights in offline environments is a risk regarding security. Because of this, Nevion wanted a security review to shed light on the security aspects of the developed concept. The total solution gives Nevion a base for implementing a similar system and utilizing this report for further research and development
Explainable AI for RGB to HyperSpectral CNN Models
No full textHyperSpectral Imaging (HSI) is a vital tool to many industries and fields. It is
however very costly, time consuming, and in need of dedicated hardware. Lots of
research was dedicated to find alternatives to traditional HSI systems. One of the
most promising ones is RGB to Hyperspectral reconstruction. These models are
usually CNNs that take in a single RGB image and estimates the hyperspectral
image for the same scene (in the visible range). Such models can dramatically cut
on costs and time needed to acquire a hyperspectral image given the availability
and ease of acquiring RGB images.
However, to fully adopt such models we need to establish trust in them (or
distrust). To do that, we need to understand and explain how these models work on
a fundamental level at least. This is especially the case because these models deal
with a highly ill-posed problem of mapping only 3 RGB bands into a much larger
number of bands (typically 31) to perform this estimation. Users do not have any
evidence of how these models actually do that and how they are able to estimate
the illuminant of the scene to avoid metameric effects and how they perform the
’one-to-many’ mapping involved. In this thesis, we work on filling this major gap.
We take 7 of the most prominent RGB to hyperspectral reconstruction models and
apply many explainable AI (XAI) methods to try to understand how they work. We
classify these models based on the different ways they perform the reconstruction.
We establish points of failure where some or all models cannot perform as expected.
We establish their spatial feature area in the input image. We try to find what
kind of parameters and features they use and where in the network they use them.
We present a theory on how they do illuminant estimation and present supporting
evidences for that theory. Finally, we bring all tests together and try to break
down these models into more simple sub-models that could be replicated by simpler
explainable equivalents. We also introduce novel modifications to existing XAI
methods that allows them to be used in any hyperspectral model explainability
project in the future.
The outcomes of this work support that these models work in an intelligible
manner. Meaning that they could be understood and equated by other explainable
models. However, these models cannot be trusted all the time, since the work
shows that they fail consistently under certain conditions. This work does not fully
explain these models since some aspects are still unclear, but it does explain many
important parts and paves the way for a clearer understanding of these networks
License Management in Closed Offline Networks Using Modern Cryptographic Solutions
No full textNevion ønsket en løsning til et system som benytter seg av en tidsbasert lisensieringsmodell der kunden selv skal kunne generere del-lisenser av en større lisens på Nevions vegne, uten internettilkobling. Gjennom dette prosjektet har det blitt utviklet et konsept til det ønskede systemet ved å benytte seg av ulike teknologier som X.509, AES-256 og PKI. Videre har det også blitt implementert et bevis på at dette konseptet er mulig å gjennomføre i praksis. Håndtering av lisenser og digitale rettigheter i miljøer uten internettilkobling innebærer en stor sikkerhetsrisiko. Derfor ønsket Nevion å vite om et slikt system er sikkert, og etterspurte en sikkerhetsanalyse av det utviklede konseptet. Den totale løsningen gir Nevion et godt utgangspunkt til å implementere et liknende system og benytte prosjektrapporten til videre forskning og utvikling.Nevion wanted a concept solution to a system that incorporates a time-based licensing model, where the customer can generate sublicenses, from a larger top level license, on the behalf of Nevion. This should happen without an internet connection. Through this project, there has been developed a concept to the wanted solution through combining technologies like X.509, AES-256 and PKI. Furthermore, there have been developed and implemented a proof of concept, proving the practicality aspect of the system. Managing licenses and digital rights in offline environments is a risk regarding security. Because of this, Nevion wanted a security review to shed light on the security aspects of the developed concept. The total solution gives Nevion a base for implementing a similar system and utilizing this report for further research and development
GUI 4 deep-doLCE
No full textHensikten med dette prosjekter er å utvikle GUI for en dyp lærings
film og bilde fargerekonstruksjon programvare kalt deep-doLCE.
Vår programvare må være forståelig og brukbare for alle med som
har skannet linseformede filmer, uansett filmarkiverings kunnskap.
Vi undersøker en enkel og intuitiv GUI som leder brukeren til neste
knapp ved å skjule og vise knappene brukeren skal trykke på
Explainable Artificial Intelligence for Image Quality Assessment
No full textImage quality assessment has been an active research field for decades because
of the high demand for images and video content in daily life. As visual information
is processed in various steps from acquisition and storage to transmission, they
are often degraded by multiple types of distortions. It is necessary to evaluate the
quality of any imaging system to maintain the user’s experience. Thus, objective
image quality assessments were proposed to objectively evaluate the image quality
as close to the perceptual quality rated by human users.
Among the three types of image quality assessment, No-Reference image quality
assessment (NR-IQA) has the most potential to be used in various applications and
is also the most challenging topic. The traditional NR-IQA metrics were proposed
using domain knowledge of natural images to extract hand-crafted features that
can indicate the degradation degree of the distorted image. Recently, many deep
learning models have been used in NR-IQA and outperform the traditional method
in predicting image quality. However, they are still data-driven models which
contain numerous parameters and lack explainability. Therefore, it is challenging
to understand how such deep NR-IQA models estimate the quality of images and
why they do not work on some images. Moreover, although many different methods
of explaining a deep learning model have been introduced, there is no work that
targets to image quality assessment.
In this work, we address the research gap in the explanation for the deep NRIQA model. Firstly, we defined a set of definitions and expectations for explainable
artificial intelligence (XAI) in the field of image quality assessment. Then, we
proposed a framework to provide explanations at different levels: from global to
local prediction for the model. The global explanations were formed by analyzing
the images that the model can not predict their quality accurately. To find such
an image, we proposed to use objective detection methods for IQA models. We
also used different existing XAI methods to obtain explanations for the model in
different information domains from spatial, and frequency to color space.
Different explanation results are discussed in our project. We found out that the
existing XAI methods can explain NR-IQA models to some extent. However, there is
no current way to evaluate the effectiveness of those explanations for image quality
assessment problems. Future work is needed to provide an objective evaluation
of XAI for image quality assessment or to find an alternative method to better
explain NR-IQA models
Réduction de dimensionalité et saillance pour la visualisation d'images spectrales
No full textNowadays, digital imaging is mostly based on the paradigm that a combinations of a small number of so-called primary colors is sufficient to represent any visible color. For instance, most cameras usepixels with three dimensions: Red, Green and Blue (RGB). Such low dimensional technology suffers from several limitations such as a sensitivity to metamerism and a bounded range of wavelengths. Spectral imaging technologies offer the possibility to overcome these downsides by dealing more finely withe the electromagnetic spectrum. Mutli-, hyper- or ultra-spectral images contain a large number of channels, depicting specific ranges of wavelength, thus allowing to better recover either the radiance of reflectance of the scene. Nevertheless,these large amounts of data require dedicated methods to be properly handled in a variety of applications. This work contributes to defining what is the useful information that must be retained for visualization on a low-dimensional display device. In this context, subjective notions such as appeal and naturalness are to be taken intoaccount, together with objective measures of informative content and dependency. Especially, a novel band selection strategy based on measures derived from Shannon’s entropy is presented and the concept of spectral saliency is introduced.De nos jours, la plupart des dispositifs numériques d’acquisition et d’affichage d’images utilisent un petit nombre de couleurs dites primaires afin de représenter n’importe quelle couleur visible. Par exemple, la majorité des appareils photos "grand public" quantifient la couleur comme une certaine combinaison de Rouge, Vert et Bleu(RVB). Ce genre de technologie est qualifiée de tri-chromatique et, au même titre que les modèles tetra-chromatiques communs en imprimerie, elle présente un certain nombre d’inconvénients, tels que le métamérisme ou encore la limitation aux longueurs d’onde visibles. Afin de palier à ces limitations, les technologies multi-, hyper,voire ultra-spectrale ont connu un gain notable d’attention depuis plusieurs décennies. Un image spectrale est constituée d’un nombre de bandes (ou canaux) supérieur à 3, représentant des régions spectrales spécifiques et permettant de recouvrer la radiance ou reflectance d’objets avec plus de précision et indépendamment du capteur utilisé. De nombreux travaux de recherche ont fait considérablement progresser les méthodes d’acquisition et d’analyse, mais beaucoup de challenges demeurent, particulièrement en ce qui concernel a visualisation de ce type de données. En effet, si une image contient plusieurs dizaines de canaux comment la représenter sur un écran qui n’en accepte que trois ? Dans cette thèse, nous présentons un certain nombre de méthodes d’extraction d’attributs pour l’analyse d’images spectrales, avec une attention particulière sur la problématique de la visualisation
Réduction de dimensionalité et saillance pour la visualisation d'images spectrales
No full textNowadays, digital imaging is mostly based on the paradigm that a combinations of a small number of so-called primary colors is sufficient to represent any visible color. For instance, most cameras usepixels with three dimensions: Red, Green and Blue (RGB). Such low dimensional technology suffers from several limitations such as a sensitivity to metamerism and a bounded range of wavelengths. Spectral imaging technologies offer the possibility to overcome these downsides by dealing more finely withe the electromagnetic spectrum. Mutli-, hyper- or ultra-spectral images contain a large number of channels, depicting specific ranges of wavelength, thus allowing to better recover either the radiance of reflectance of the scene. Nevertheless,these large amounts of data require dedicated methods to be properly handled in a variety of applications. This work contributes to defining what is the useful information that must be retained for visualization on a low-dimensional display device. In this context, subjective notions such as appeal and naturalness are to be taken intoaccount, together with objective measures of informative content and dependency. Especially, a novel band selection strategy based on measures derived from Shannon’s entropy is presented and the concept of spectral saliency is introduced.De nos jours, la plupart des dispositifs numériques d’acquisition et d’affichage d’images utilisent un petit nombre de couleurs dites primaires afin de représenter n’importe quelle couleur visible. Par exemple, la majorité des appareils photos "grand public" quantifient la couleur comme une certaine combinaison de Rouge, Vert et Bleu(RVB). Ce genre de technologie est qualifiée de tri-chromatique et, au même titre que les modèles tetra-chromatiques communs en imprimerie, elle présente un certain nombre d’inconvénients, tels que le métamérisme ou encore la limitation aux longueurs d’onde visibles. Afin de palier à ces limitations, les technologies multi-, hyper,voire ultra-spectrale ont connu un gain notable d’attention depuis plusieurs décennies. Un image spectrale est constituée d’un nombre de bandes (ou canaux) supérieur à 3, représentant des régions spectrales spécifiques et permettant de recouvrer la radiance ou reflectance d’objets avec plus de précision et indépendamment du capteur utilisé. De nombreux travaux de recherche ont fait considérablement progresser les méthodes d’acquisition et d’analyse, mais beaucoup de challenges demeurent, particulièrement en ce qui concernel a visualisation de ce type de données. En effet, si une image contient plusieurs dizaines de canaux comment la représenter sur un écran qui n’en accepte que trois ? Dans cette thèse, nous présentons un certain nombre de méthodes d’extraction d’attributs pour l’analyse d’images spectrales, avec une attention particulière sur la problématique de la visualisation
GUI 4 deep-doLCE
No full textHensikten med dette prosjekter er å utvikle GUI for en dyp lærings
film og bilde fargerekonstruksjon programvare kalt deep-doLCE.
Vår programvare må være forståelig og brukbare for alle med som
har skannet linseformede filmer, uansett filmarkiverings kunnskap.
Vi undersøker en enkel og intuitiv GUI som leder brukeren til neste
knapp ved å skjule og vise knappene brukeren skal trykke på.The purpose of the project is to develop a Graphical User Interface for a
deep learning film and image color reconstruction software called
deep-doLCE. The application needs to be understandable and usable by
anyone with a scanned lenticular film, regardless of any film archiving
knowledge or not. We propose a simple and intuitive GUI that guides the
user to their next appropriate action by hiding and showing the buttons
they should press next
Réduction de dimensionalité et saillance pour la visualisation d'images spectrales
No full textNowadays, digital imaging is mostly based on the paradigm that a combinations of a small number of so-called primary colors is sufficient to represent any visible color. For instance, most cameras usepixels with three dimensions: Red, Green and Blue (RGB). Such low dimensional technology suffers from several limitations such as a sensitivity to metamerism and a bounded range of wavelengths. Spectral imaging technologies offer the possibility to overcome these downsides by dealing more finely withe the electromagnetic spectrum. Mutli-, hyper- or ultra-spectral images contain a large number of channels, depicting specific ranges of wavelength, thus allowing to better recover either the radiance of reflectance of the scene. Nevertheless,these large amounts of data require dedicated methods to be properly handled in a variety of applications. This work contributes to defining what is the useful information that must be retained for visualization on a low-dimensional display device. In this context, subjective notions such as appeal and naturalness are to be taken intoaccount, together with objective measures of informative content and dependency. Especially, a novel band selection strategy based on measures derived from Shannon’s entropy is presented and the concept of spectral saliency is introduced.De nos jours, la plupart des dispositifs numériques d’acquisition et d’affichage d’images utilisent un petit nombre de couleurs dites primaires afin de représenter n’importe quelle couleur visible. Par exemple, la majorité des appareils photos "grand public" quantifient la couleur comme une certaine combinaison de Rouge, Vert et Bleu(RVB). Ce genre de technologie est qualifiée de tri-chromatique et, au même titre que les modèles tetra-chromatiques communs en imprimerie, elle présente un certain nombre d’inconvénients, tels que le métamérisme ou encore la limitation aux longueurs d’onde visibles. Afin de palier à ces limitations, les technologies multi-, hyper,voire ultra-spectrale ont connu un gain notable d’attention depuis plusieurs décennies. Un image spectrale est constituée d’un nombre de bandes (ou canaux) supérieur à 3, représentant des régions spectrales spécifiques et permettant de recouvrer la radiance ou reflectance d’objets avec plus de précision et indépendamment du capteur utilisé. De nombreux travaux de recherche ont fait considérablement progresser les méthodes d’acquisition et d’analyse, mais beaucoup de challenges demeurent, particulièrement en ce qui concernel a visualisation de ce type de données. En effet, si une image contient plusieurs dizaines de canaux comment la représenter sur un écran qui n’en accepte que trois ? Dans cette thèse, nous présentons un certain nombre de méthodes d’extraction d’attributs pour l’analyse d’images spectrales, avec une attention particulière sur la problématique de la visualisation
Towards Understanding Change Blindness Based on Cognitive Attributes and Stimulus Complexity
No full textThe properties of the human visual system such as limited working memory and attention can cause human observers to miss out on otherwise very conspicuous changes in their visual field. This phenomenon, known as Change Blindness (CB), can create risk when a human observer is interacting with interfaces using visual changes to communicate important information. As a result, the prediction of the probability of CB occurrence has gained importance. Although numerous studies have explored the change blindness phenomena, few of them have explored the connection between CB and individual characteristics. Additionally, many studies employ natural scenes to study change blindness which are too complex to be used for the fundamental modeling of CB.
In this work, the connection between the CB parameters such as CB performance or response time(RT) with several cognitive parameters such as distractibility, cognitive failure, the performance of visual short-term memory, and the robustness of perception in the presence of confusing signals were explored. However, instead of employing natural image stimuli with high semantic content, crafted, well-controlled stimuli involving less semantic content were used. The stimuli were composed of patterns caused by variations in attributes of the included elements such as color, shape, or orientation, and cases with more than one attribute were also generated. Additionally, eye-tracking was used to explore the information revealed by the observers' eye movement.
The results revealed the absence of a connection between the attention-related parameters and CB parameters in the present setup. On the other hand, the results suggested a significant correlation between the CB parameters and the visual short-term memory (VSTM) parameters. This influence was further approved by the observed correlations between the VSTM parameters and the number of cases where the observer looked at the change yet failed to perceive it.The properties of the human visual system such as limited working memory and attention can cause human observers to miss out on otherwise very conspicuous changes in their visual field. This phenomenon, known as Change Blindness (CB), can create risk when a human observer is interacting with interfaces using visual changes to communicate important information. As a result, the prediction of the probability of CB occurrence has gained importance. Although numerous studies have explored the change blindness phenomena, few of them have explored the connection between CB and individual characteristics. Additionally, many studies employ natural scenes to study change blindness which are too complex to be used for the fundamental modeling of CB.
In this work, the connection between the CB parameters such as CB performance or response time(RT) with several cognitive parameters such as distractibility, cognitive failure, the performance of visual short-term memory, and the robustness of perception in the presence of confusing signals were explored. However, instead of employing natural image stimuli with high semantic content, crafted, well-controlled stimuli involving less semantic content were used. The stimuli were composed of patterns caused by variations in attributes of the included elements such as color, shape, or orientation, and cases with more than one attribute were also generated. Additionally, eye-tracking was used to explore the information revealed by the observers' eye movement.
The results revealed the absence of a connection between the attention-related parameters and CB parameters in the present setup. On the other hand, the results suggested a significant correlation between the CB parameters and the visual short-term memory (VSTM) parameters. This influence was further approved by the observed correlations between the VSTM parameters and the number of cases where the observer looked at the change yet failed to perceive it
- …
