1,720,992 research outputs found
Automatically predicting mood from expressed emotions
No full textAffect-adaptive systems have the potential to assist users that experience systematically negative moods. This thesis aims at building a platform for predicting automatically a person’s mood from his/her visual expressions. The key word is mood, namely a relatively long-term, stable and diffused affective state, as opposed to the short-term, volatile and intense emotion. This is emphasized, because mood and emotion often tend to be used as synonyms. However, since their differences are well established in the psychological literature, to address the thesis objective, we consider mood recognition as a different problem from emotion recognition. The main and key idea is to discover whether using the expressed emotions of a person over time can help us estimate the mood. The advantage of this notion is that we can reuse the bulk of work on automatic emotion recognition and plug in the mood recognition module. A big part of this thesis focuses on unveiling a functional relationship between the expressed emotions and mood. We first set out to verify whether simple aggregation rules, such as the average emotion, are good approximation of the mood. We continue by building incrementally more complex models that fit better the peculiarities of the mood estimation function. Equally important to retrieving the mood function, is to validate it on the proper data. We dedicate special attention on crafting a database that contains sufficiently long videos to capture the mood of a person, expressed subtly through genuine expressions. We annotate the expressed moods in the videos through crowdsourcing, as a fast way to obtain multiple ground truth labels per video. Our devised mood model and the publicly available database set a strong basis towards the development of the envisioned affect-assistive system, and more broadly towards the research on automatic mood recognition from videos.Intelligent SystemsElectrical Engineering, Mathematics and Computer Scienc
Evaluating the impact of digital filters on the aesthetic appeal of photographs: A crowdsourcing based approach
No full textIn particular grouping by aesthetics and quality of the media has brought along new challenges for Computational Aesthetics research such as what makes an image beautiful, what means beautiful and how do you quantify beautiful. to meet those challenges, researchers have tried to come up with several algorithms based in different metrics to bridge the gap between the quantitative aspects of what is called beauty and what people call beauty. In order to fill part of this gap we studied the effect of digital filters in photographic aesthetics so widely used in the social networks nowadays. Taking in consideration the popularity of digital filters among many social network users, it was a surprise to understand that most participants in the experiment preferred the images with no filter. In any case measuring what is beautiful always requires collecting aesthetics scores from people. Doing that collection process in a laboratory environment is the most effective approach. The main reasons are the highly controlled environment that leads to good data quality. The downside is cost, time and restriction of participants to the people available nearby. Therefore another issue addressed in the study was the use of crowdsourcing to minimize time and cost, as well as to expand the scope of participation, in the process of collecting image scores from users. To test that possibility a 4 step process step was designed and implemented. First preference scores were collected in a lab environment over a previous selected dataset. Afterwards the crowdsourcing experiment was planned what included an optimization of the dataset (ground truth dataset). Subsequently three digital filters were then applied to the collection and an online experiment followed to once again collect preference scores. In phase, we developed the experiment in the context of Microworkers and as a Facebook app interface enriched with a playful visual interface. The last step included a process to filter the suspicious participants and check results consistency. The results show that implementing an experiment to collect preferences of image quality in social media is a good methodology for Computational Aesthetics, if appropriate planning and management is adopted.Embedded SystemsInteractive Intelligent Systems GroupElectrical Engineering, Mathematics and Computer Scienc
No-Reference Image Quality Assessment using Deep Convolutional Neural Networks
No full textNo-reference image quality assessment (NR-IQA) is a challenging field of research that, without making use of reference images, aims at predicting the image quality as it is perceived by the human visual system (HVS). Many NR-IQA methods have been proposed over time but recently proposed convolutional neural network (CNN) based approaches, through their powerful feature learning capabilities, have outperformed all previously existing NR-IQA methods. But these CNN based approaches are perceptually incorrect in assuming distortions to be homogeneously distributed across images. They operate on very small image portions while considering all of them to have identical perceptual quality, whereas in reality, different parts of an image, based on their structure and content, could bear different perceptual quality. Further, these approaches utilize shallow CNN architectures which render them incapable of taking advantages offered by the deep CNN architectures. To improve upon the limitations of existing CNN based approaches, we conducted a design space exploration of CNN’s and proposed a suitable CNN design for NR-IQA task, that operates on bigger image portions and employs a deeper architecture. The proposed design achieves the state of the art performance on LIVE and TID datasets. We further provide informative visualization of features learned by the proposed CNN design, which shed light on its internal working while promoting further understanding regarding the nature of image quality.Electrical Engineering, Mathematics and Computer ScienceIntelligent System
Creating a Mood Database for automated affect analysis
No full textAffect-adaptive systems are dependent on their ability to automatically recognize a user’s affective state. This study aims to contribute to the creation of an affect-adaptive system that can recognize negative moods of elderly in care homes from a video feed, and improve it by adapting the lighting in the room. An affective database of videos portraying different moods is required to train such a system. While many affective databases exist already, they are primarily targeting emotions rather than mood. Therefore, we introduce a new database of annotated videos that can be used for mood recognition. To maintain control over which moods are depicted in the videos in the database, we combine the use of mood induction and acted performance to portray the moods in a realistic way, incorporating in the acted scripts the results from a series of interviews with caretakers in care homes. The database covers three visual modalities: body, face and 3D Kinect data for a total of 24 hours of recorded video material. We use crowdsourcing to annotate such a large amount of material in terms of perceived mood of the person portrayed in the videos, by outsourcing via the internet the annotation task to a large number of paid annotators. A risk of using crowdsourcing is unreliable annotator performance, due to the low level of control applicable to the annotation process. We deal with this problem by filtering the annotations according to predefined criteria, checking for task commitment and self-consistency of the annotators. We validate our use of the combination of induction and actors with a comparison between the intended mood, the mood felt by the actors, and the mood perceived by annotators. Furthermore, we demonstrate that crowdsourcing is a promising tool for the annotation of mood.Electrical Engineering, Mathematics and Computer ScienceIntelligent SystemsInteractive Intelligenc
1Mbps is enough: video quality and individual idiosyncrasies in multiparty HD video-conferencing
Full text linkMost video platforms deliver HD video in high bitrate encoding. Modern video-conferencing systems are capable of handling HD streams, but using multiparty conferencing, average internet connections in the home are on their bandwidth limit. For properly managing the encoding bitrate in videoconferencing, we must know what is the minimum bitrate requirement to provide users an acceptable experience, and what is the bitrate level after which QoE saturates?. Most available subjective studies in this area used rather dated technologies. We report on a multiparty study on video quality with HD resolution. We tested different encoding bitrates (256kbs, 1024kbs and 4096kbs) and packet loss rates (0, 0.5%) in groups of 4 participants with a scenario based on the ITU building blocks task. We discuss the influence of group interaction and individual idiosyncrasies based on different mixed models, and look at covariates engagement and enjoyment as further explanatory factors. We found that 256kbs is still sufficient to provide a fair overall experience, but video quality is noticed to be poor. On the higher bitrate end, most people will not perceive the difference between 1024kbs and 4096kbs, considering in both cases the quality to be close to excellent. Independent on bitrate, packet loss has a small but significant impact, quantifiable in, on average, less than half a point difference on a 5-point ITU scale
Transfer Learning for Rain Detection in Images
No full textExtreme weather conditions seem to occur stronger and more frequently due to climate change. Very expensive technology is used to predict them, which results problematic when they have to be used in less developed countries. An alternative could be to employ digital sensors, such as phone cameras or existing webcam infrastructures, widely distributed in many countries in the world, and to analyse the captured images. Many methods have been proposed for weather detection including also Convolutional Neural Networks (CNNs). The latter have recently become very popular in the field of computer vision due to their excellent performance in image recognition tasks. However, CNNs are characterized by a high number of learnable parameters that need an equally high number of data points to achieve good performance. Since very big dataset are hardly available, techniques that help to overcome this problem, such as transfer learning, can be used. Different are the transfer learning approaches: the fine-tuning approach, i.e. re-training all the network layers, and the freezing layers approach, i.e. re-training just a subset of the network layers. In collaboration with IBM Netherlands Center for Advance Studies, we optimized a ResNet-18 architecture, modifying the architecture depth and applying regularization methods, to perform weather detection of images showing rain or no-rain conditions. The architecture was previously trained on the ImageNet dataset and then, through the fine-tuning approach, we re-trained the network layers using as training data points weather images captured by webcams distributed in The Netherland, Belgium and London. In particular, we collected a dataset composed by 397041 images showing scenarios such as city roads, urban and rural areas. We also adopted the freezing layer approach on an optimized ResNet-18 architecture and made comparison between the two approaches in relation to weather detection task.Electrical Engineering, Mathematics and Computer ScienceComputer Scienc
Worker communities in online crowdsourcing markets
No full textElectrical Engineering, Mathematics and Computer ScienceSoftware Technolog
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
On the influences of personality traits on employees engagement with gamified enterprise tools
No full textGamification techniques are used in enterprises to support employees' engagement with computer-mediated business processes. The potential effectiveness of the incentives brought by gamification techniques are, however, not equally appealing to individuals. To better understand when gamification can be an effective engagement aid, it is important to study how individual differences (personal or character-related) of employees relate with the effectiveness of game mechanics applied to enterprise-class computer tools. Personality is a property of an individual that is known to influence, among others, task performance, learning styles, and gaming preferences. Despite the existence of an abundant body of research, the relationship between the effectiveness of game mechanics in an enterprise setting and the personality of employees is yet to be fully understood. This thesis contributes new knowledge on the matter, by studying the influence of personality traits and gender stereotypes on the behavior of 177 IBM employees that participated in an experiment on gamified learning and social experience. We engaged with the employees of IBM Netherlands and performed a personality trait and gender stereotype inventory by means of a questionnaire. The results of the questionnaire supported our investigation on the descriptive power of personality traits in explaining the differences in participation and engagement in the targeted population. Finally, we validated the effectiveness of state-of-the-art techniques for automated personality assessment, to assess the possibility of developing large-scale experiments on the effect personality traits without the need for questionnaires.Master of Computer ScienceWeb Information SystemsElectrical Engineering, Mathematics and Computer Scienc
- …
