Search

We present a novel solution to the problem of detecting common actions in time series of motion capture data and videos. Given two action sequences, our method discovers all pairs of common subsequences, i.e. subsequences that represent the same or similar action. This is achieved in a completely unsupervised manner, i.e., without any prior knowledge of the type of actions, their number and their duration. These common subsequences (commonalities) may be located anywhere in the original sequences, may differ in duration and may be performed under different conditions e.g., by a different actor. The proposed method performs a very efficient graph-based search on the matrix of pairwise distances of frames of the two sequences. This search is supported by an objective function that captures the trade off between the similarity of the common subsequences and their lengths. The proposed method has been evaluated quantitatively on challenging datasets and in comparison to state of the art approaches. The obtained results demonstrate that the proposed method outperforms the state of the art methods both in the quality of the obtained solutions and in computational performance. [ABSTRACT FROM AUTHOR]

Recommender systems try to predict the preferences of users for specific items, based on an analysis of previous consumer preferences. In this paper, we propose SCoR, a Synthetic Coordinate based Recommendation system which is shown to outperform the most popular algorithmic techniques in the field, approaches like matrix factorization and collaborative filtering. SCoR assigns synthetic coordinates to nodes (users and items), so that the distance between a user and an item provides an accurate prediction of the user’s preference for that item. The proposed framework has several benefits. It is parameter free, thus requiring no fine tuning to achieve high performance, and is more resistance to the cold-start problem compared to other algorithms. Furthermore, it provides important annotations of the dataset, such as the physical detection of users and items with common and unique characteristics as well as the identification of outliers. SCoR is compared against nine other state-of-the-art recommender systems, sever of them based on the well known matrix factorization and two on collaborative filtering. The comparison is performed against four real datasets, including a brief version of the dataset used in the well known Netflix challenge. The extensive experiments prove that SCoR outperforms previous techniques while demonstrating its improved stability and high performance. [ABSTRACT FROM AUTHOR]

In this work, an unsupervised isocontour based segmentation method is proposed, that is applied on the detection of topographic highs with arbitrary basal shapes on Digital Elevation Models (DEMs). A series of isocontour based segmentation maps is computed for decreasing altitude levels. During this process, the isocontours are gradually merged providing a topological hierarchy of highs in an inclusion tree structure. A novel formulation of a topographic high is given taking into account the volume evolution of an isocontour that starts from the top of a high and grows, as decreasing the altitude level of isocontour, until a high of higher altitude is reached. This formulation yields to a robust unsupervised algorithm that can be sequentially applied to automatically recognize and discriminate the topographic highs of a region according to the inclusion tree without any constraint on basal shapes. The proposed method is applied on real and synthetic DEMs, in order to automatically detect the exact shape of complex topographic highs and some geomorphological based features useful for high annotations, yielding high performance results, even if the highs are partially visible in the given DEM. [ABSTRACT FROM AUTHOR]

Our goal is to represent a given 2D shape with an automatically determined number of ellipses, so that the total area covered by the ellipses is equal to the area of the original shape without any assumption or prior knowledge about the object structure. To solve this interesting theoretical problem, first we employ the skeleton of the 2D shape which provides important information on the parameters of the ellipses that could approximate the original shape. For a given number of such ellipses, the hard Expectation-Maximisation (EM) algorithm is employed to maximise the shape coverage under the Equal Area constraint. Different models (i.e., solutions involving different numbers of ellipses) are evaluated based on the Akaike Information Criterion (AIC). This considers a novel, entropy-based shape complexity measure that balances the model complexity and the model approximation error. In order to minimise the AIC criterion, two variants are proposed and evaluated: (a) the augmentative method that gradually increases the number of considered ellipses starting from a single one and (b) the decremental method that decreases the number of ellipses starting from a large, automatically defined set. The obtained quantitative results on more than 4000 2D shapes included in standard as well as in custom datasets, quantify the performance of the proposed methods and illustrate that their solutions agree with human intuition. [ABSTRACT FROM AUTHOR]

Oil spill models are combined with bathymetric, meteorological, oceanographic, and geomorphological data to model a series of oil spill accidents in the Eastern Mediterranean Sea. A total of 104 oil spill simulations, computed for 11 different locations in the Levantine Basin, show that oil slicks will reach the coast of Cyprus in four (4) to seven (7) days in summer conditions. Oil slick trajectories are controlled by prevailing winds and current eddies. Based on these results, we support the use of chemical dispersants in the very few hours after large accidental oil spills. As a corollary, we show shoreline susceptibility to vary depending on: a) differences in coastline morphology and exposure to wave action, b) the existence of uplifted wave-cut platforms, coastal lagoons and pools, and c) the presence of tourist and protected environmental areas. Mitigation work should take into account the relatively high susceptibility of parts of the Eastern Mediterranean. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, we propose a framework for interactive image segmentation. The goal of interactive image segmentation is to classify the image pixels into foreground and background classes, when some foreground and background markers are given. The proposed method minimizes a min–max Bayesian criterion that has been successfully used on image segmentation problem and it consists of several steps in order to take into account visual information as well as the given markers, without any requirement of training. First, we partition the image into contiguous and perceptually similar regions (superpixels). Then, we construct a weighted graph that represents the superpixels and the connections between them. An efficient algorithm for graph clustering based on synthetic coordinates is used yielding an initial map of classified pixels. This method reduces the problem of graph clustering to the simpler problem of point clustering, instead of solving the problem on the graph data structure, as most of the known algorithms from literature do. Finally, having available the data modeling and the initial map of classified pixels, we use a Markov Random Field (MRF) model or a flooding algorithm to get the image segmentation by minimizing a min–max Bayesian criterion. Experimental results and comparisons with other methods from the literature are presented on LHI, Gulshan and Zhao datasets, demonstrating the high performance and accuracy of the proposed scheme. [Copyright &y& Elsevier]

We propose a method to improve the prediction performance of recommender systems via a Dual (user anditem) Training Error based Correction approach (DTEC). The proposed method is applied to the Synthetic Coordinate Recommendation system (SCoR) (Papadakis et al., 2017) and to other Ithree state-of-the-art systems. Initially, a recommender system is used Ito provide recommendations for users and items. Subsequently, we introduce a second stage, after initial execution of the recommender system, that improves its predictions taking into account the error in the training set between users and items and their similarity. These corrections can be performed from both user and item viewpoints, and finally a dual system is proposed that efficiently combines both corrections. DTEC computes a model that makes zero the recommendation error in the training set, and then applies it on the test set to improve the rating predictions. The proposed DTEC approach is applicable Ito any model-based recommender system with positive training error, potentially increasing the accuracy of the recommendations. The experimental results demonstrate the efficiency and high performance of DTEC on four well-known, real-world datasets. • DTEC takes into account the error in the training set between users and items. • A dual system efficiently combines the user and item view point corrections. • Improvement of recommendations of the SCoR system. • Applicability of DTEC on other recommender systems to improve recommendations. • High performance results on four well-known, real world datasets. [ABSTRACT FROM AUTHOR]