Overview and details of the sessions of this conference. Please register as a participant for the conference (free!) and then Login in order to have access to downloads in the detailed view. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view.
IVA3-P: Image and Video Analysis 3
5:40pm - 5:45pm
Optimizing Video Quality Estimation Across Resolutions
University of Texas at Austin, United States of America
Many algorithms have been developed to evaluate the perceptual quality of images and videos, based on models of picture statistics and visual perception. These algorithms attempt to capture user experience better than simple metrics like the peak signal-to-noise ratio (PSNR) and are widely utilized on streaming service platforms and in social networking applications to improve users' Quality of Experience. The growing demand for high-resolution streams and rapid increases in user-generated content (UGC) sharpens interest in the computation involved in carrying out perceptual quality measurements. In this direction, we propose a suite of methods to efficiently predict the structural similarity index (SSIM) of high-resolution videos distorted by scaling and compression, from computations performed at lower resolutions. We show the effectiveness of our algorithms by testing on a large corpus of videos and on subjective data.
5:45pm - 5:50pm
Hybrid Motion Magnification based on Same-Frame Optical Flow Computations
University of Brasilia, Brazil
Motion magnification refers to the ability of amplifying small movements in a video in order to reveal important information about the observed scene. In the past, several motion magnification methods have been proposed, but most of them have the disadvantage of introducing annoying visual artifacts in the video. In this paper, we propose a method that analyses the optical flow between each original frame and the corresponding motion-magnified frame and, then, synthesizes a new motion-magnified video by remapping the original video using the generated optical flow map. When compared to state-of-the-art motion magnification methods, the proposed approach is able to achieve a higher motion magnification, without introducing strong visual artifacts.
5:50pm - 5:55pm
On Maximum A Posteriori Approximation of Hidden Markov Models for Proportional Data
Concordia University, Canada
Hidden Markov models (HMM) have recently risen as a key generative machine learning approach for time series data study and analysis. While early works focused only on applying HMMs for speech recognition, HMMs are now prominent in various fields such as video classification and genomics. In this paper, we develop a Maximum A Posteriori framework for learning the Generalized Dirichlet HMMs that have been proposed recently as an efficient way for modeling sequential proportional data. In contrast to the conventional Baum Welch algorithm, commonly used for learning HMMs, the proposed algorithm places priors for the learning of the desired parameters; hence, regularizing the estimation process. We validate our proposed approach on a challenging video processing application; namely, dynamic texture classification.
5:55pm - 6:00pm
A Non-Negative Matrix Factorization Framework for Privacy-Preserving and Federated Learning
Oregon State University, United States of America
The uncontrolled growth in domains such as a surveillance system, health care, and finance produce a large amount of data and contain potentially sensitive data that can become public if they are not appropriately sanitized.
Motivated by this issue, we introduce a privacy filter (PF), a novel non-negative matrix factorization (NMF) framework which aims to preserve the privacy of data before publishing.
More specifically, this framework enables data holders to choose the data dimension that protects user privacy without being aware of the privacy leakage.
Also, we consider the problem of privately learning a PF across multiple sensitive datasets, leading to a federated learning algorithm guaranteeing the protection of private data and high accuracy classification for non-private information.
Finally, the experiments conduct and illustrate the superior performance of the proposed algorithms under the premise of protecting users’ private data.