Invited SpeakersProfessor Jiebo Luo
Department of Computer Science
University of Rochester
Computational Inference of Emotion in Images: A Final Frontier and a Data-Intensive Approach
With the recent successes in using deep learning techniques to solve computer vision problems, the performances of the state of the art algorithms in many areas, especially object recognition, have been dramatically improved. Researchers are now inspired to address yet more challenging problems, such as associating pictures with aesthetics, and have also reported progress. One remaining final frontier in extracting meaning from images is related to the recognition of emotions that images arouse in humans. This is one area where the traditional approaches with low-level image features have so far failed to make any inroads. The key challenges are the loose and highly abstract nature of semantics associated with emotions. Is it possible to tackle such a complex problem using a data-intensive approach? In addition to being a challenging research topic, sentiment or emotion analysis from images has become an important issue with practical implications. Today, social networks have grown to be perhaps the most important sources for people to acquire information on all aspects of their lives. Meanwhile, every online social network user is a contributor to such large amounts of information. Online users love to share their experiences and to express their opinions on virtually all events and subjects. The large amount of online user generated data convey people’s opinions or sentiments towards specific topics and events. There have been many works on using online users’ sentiments to make predictions in many domains, yet most of them only rely on sentiment analysis from textual content. How to reliably extract emotion signals from images and multimedia? We will discuss how to effectively employ a data-intensive approach to emotion recognition in images, as well as multimedia that include both image and text information.
Download PDF [1.8 MB]
Professor Milan Sonka
Department of Electrical and Computer Engineering
The University of Iowa
Relating Retinal Anatomy, Pathology, Function, and Therapy Guidance: Precision Medicine via Analysis of Ophthalmic 3D OCT
Accurate and reliable image segmentation is of great importance in quantitative medical image analysis and precision medicine. In ophthalmology, translational applications of medical imaging were – until recently – limited to 2D analyses of fundus photographs. With a fast-growing routine clinical use of 3-D imaging modalities like optical coherence tomography (OCT), ophthalmologists (same as radiologists decades ago) are faced with ever-increasing amounts of image data to analyze. Quantitative outcomes of such analyses are growing in importance. Yet, daily interpretation of clinical ophthalmic OCT images is still typically performed visually and qualitatively, with quantitative clinical analysis being an exception rather than the norm. Since performing full OCT image segmentations in 3D is infeasible for a physician in clinical setting due to the time constraints, quantitative and highly automated analysis methods must be developed. Our approach to simultaneous segmentation of multiple interacting surfaces appearing in the context of other interacting objects will be presented. The reported methods are part of the family of graph-based image segmentation methods dubbed LOGISMOS for Layered Optimal Graph Image Segmentation of Multiple Objects and Surfaces. This family of methods guarantees solution optimality with direct applicability to n-D problems. The talk will present new methods and approaches developed during our ongoing ophthalmic OCT image analysis projects, including morphologic analyses of normal and pathologic retinal OCT, determination of structure—function relationships in glaucoma, methods for image-guided treatment of age-related macular degeneration, and approaches applicable to other vision impairing and/or blinding diseases.
Download PDF [7 MB]
Professor Zhou Wang
Department of Electrical and Computer Engineering
University of Waterloo
Objective Image Quality Assessment – Current Status and What's Beyond
Images and videos are subject to a wide variety of distortions during acquisition, processing, compression, transmission and reproduction. Humans can effortlessly identify image quality degradations. By contrast, objective evaluation of perceived image quality turns out to be a challenging task. In the past decade, there has been an accelerated interest in objective image quality assessment methodologies, whose roles are not only to monitor image quality degradations and to benchmark image processing systems, but also to optimize a large variety of image and video processing algorithms and systems. In this talk, we will first give a brief overview of the field of objective image and video quality assessment. We will then move forward to discuss two rapidly growing research areas that extend image quality research to a larger space of great values and broader impact. The first is to use objective perceptual image quality models as the optimization criteria in the development of novel image processing algorithms and systems. The second is to extend the concept of image quality to visual quality-of-experience (QoE) in network visual communication environment, which could reshape the future of video delivery and impact the experience of everyone who watches video through the network.
Download PDF [4 MB]
ICIAR - International Conference on Image Analysis and Recognition
Send your comments to firstname.lastname@example.org