Fractal dimension of bag-of-visual words

Pattern Analysis and Applications - Scene recognition is an important and challenging problem in computer vision. One of the most used scene recognition methods is the bag-of-visual words. Despite...     

Second-order optimization of mutual information for real-time image registration

In this paper, we present a direct image registration approach that uses mutual information (MI) as a metric for alignment. The proposed approach is robust and gives an accurate estimation of a set of 2-D motion parameters in real time. MI is a measure of the quantity of information shared by signals. Although it has the ability to perform robust alignment with illumination changes, multimodality, and partial occlusions, few works have proposed MI-based applications related to spatiotemporal image registration or object tracking in image sequences because of some optimization problems, which we will explain. In this paper, we propose a new optimization method that is adapted to the MI cost function and gives a practical solution for real-time tracking. We show that by refining the computation of the Hessian matrix and using a specific optimization approach, the registration results are far more robust and accurate than the existing solutions, with the computation also being cheaper. A new approach is also proposed to speed up the computation of the derivatives and keep the same optimization efficiency. To validate the advantages of the proposed approach, several experiments are performed.     

Material challenges for solar cells in the twenty-first century: directions in emerging technologies

Photovoltaic generation has stepped up within the last decade from outsider status to one of the important contributors of the ongoing energy transition, with about 1.7% of world electricity provided by solar cells. Progress in materials and production processes has played an important part in this development. Yet, there are many challenges before photovoltaics could provide clean, abundant, and cheap energy. Here, we review this research direction, with a focus on the results obtained within a Japan–French cooperation program, NextPV, working on promising solar cell technologies. The cooperation was focused on efficient photovoltaic devices, such as multijunction, ultrathin, intermediate band, and hot-carrier solar cells, and on printable solar cell materials such as colloidal quantum dots.     

Optics: finding the best fit in telecom transport networks

This paper aims at looking beyond the gloomy current market evolution of the optical industry. Based on an analysis of the telecom transport market evolution (metropolitan as well as backbone), taking benefit of very significant technological breakthroughs in optics in recent years, this paper will suggest promising, yet pragmatic applications in telecom transport networks where optics will play a significant role beyond pure optical transmission.     

Assessing spatio-temporal rainfall variability in a tropical mountain area (Ethiopia) using NOAA’s rainfall estimates

Seasonal and interannual variation in rainfall can cause massive economic loss for farmers and pastoralists, not only because of deficient total rainfall amounts but also because of long dry spells within the rainy season. The semi-arid to sub-humid mountain climate of the North Ethiopian Highlands is especially vulnerable to rainfall anomalies. In this article, spatio-temporal rainfall patterns are analysed on a regional scale in the North Ethiopian Highlands using satellite-derived rainfall estimates (RFEs). To counter the weak correlation in the dry season, only the rainy season rainfall from March till September is used, responsible for approximately 91% of the annual rainfall. Validation analysis demonstrates that the RFEs are well correlated with the meteorological station (MS) rainfall data, i.e. 85% for RFE 1.0 (1996–2000) and 80% for RFE 2.0 (2001–2006). However, discrepancies indicate that RFEs generally underestimate MS rainfall and the scatter around the trendlines indicates that the estimation by RFEs can be in gross error. A local calibration of RFE with rain gauge information is validated as a technique to improve RFEs for a regional mountainous study area. Slope gradient, slope aspect, and elevation have no added value in the calibration of the RFEs. The estimation of monthly rainfall using this calibration model improved on average by 8%. Based upon the calibration model, annual rainfall maps and an average isohyet map for the period 1996–2006 were constructed. The maps show a general northeast–southwest gradient of increasing rainfall in the study area and a sharp east–west gradient in its northern part. Slope gradient, slope aspect, elevation, easting, and northing were evaluated as explanatory factors for the spatial variability of annual rainfall in a stepwise multiple regression with the calibrated average of RFE 1.0 as dependent variable. Easting and northing are the only significant contributing variables (R² = 0.86), of which easting has proved to be the most important factor (R² = 0.72). The scatter around the individual trendlines of easting and northing corresponds to an increase in rainfall variability in the drier regions. Despite the remaining underestimation of rainfall in the southern part of the study area, the improved estimation of spatio-temporal rainfall variability in a mountainous region by RFEs is valuable as input to a wide range of scientific models.     

Local receptive fields based extreme learning machine with hybrid filter kernels for image classification

Multidimensional Systems and Signal Processing - In this paper, an innovative method called extreme learning machine with hybrid local receptive fields (ELM-HLRF) is presented for image...     

Residual variance estimation in machine learning

The problem of residual variance estimation consists of estimating the best possible generalization error obtainable by any model based on a finite sample of data. Even though it is a natural generalization of linear correlation, residual variance estimation in its general form has attracted relatively little attention in machine learning.In this paper, we examine four different residual variance estimators and analyze their properties both theoretically and experimentally to understand better their applicability in machine learning problems. The theoretical treatment differs from previous work by being based on a general formulation of the problem covering also heteroscedastic noise in contrary to previous work, which concentrates on homoscedastic and additive noise.In the second part of the paper, we demonstrate practical applications in input and model structure selection. The experimental results show that using residual variance estimators in these tasks gives good results often with a reduced computational complexity, while the nearest neighbor estimators are simple and easy to implement.     

Learning probabilistic models of tree edit distance

Nowadays, there is a growing interest in machine learning and pattern recognition for tree-structured data. Trees actually provide a suitable structural representation to deal with complex tasks such as web information extraction, RNA secondary structure prediction, computer music, or conversion of semi-structured data (e.g. XML documents). Many applications in these domains require the calculation of similarities over pairs of trees. In this context, the tree edit distance (ED) has been subject of investigations for many years in order to improve its computational efficiency. However, used in its classical form, the tree ED needs a priori fixed edit costs which are often difficult to tune, that leaves little room for tackling complex problems. In this paper, to overcome this drawback, we focus on the automatic learning of a non-parametric stochastic tree ED. More precisely, we are interested in two kinds of probabilistic approaches. The first one builds a generative model of the tree ED from a joint distribution over the edit operations, while the second works from a conditional distribution providing then a discriminative model. To tackle these tasks, we present an adaptation of the expectation–maximization algorithm for learning these distributions over the primitive edit costs. Two experiments are conducted. The first is achieved on artificial data and confirms the interest to learn a tree ED rather than a priori imposing edit costs; The second is applied to a pattern recognition task aiming to classify handwritten digits.     

Methodology for long-term prediction of time series

In this paper, a global methodology for the long-term prediction of time series is proposed. This methodology combines direct prediction strategy and sophisticated input selection criteria: k-nearest neighbors approximation method (k-NN), mutual information (MI) and nonparametric noise estimation (NNE). A global input selection strategy that combines forward selection, backward elimination (or pruning) and forward–backward selection is introduced. This methodology is used to optimize the three input selection criteria (k-NN, MI and NNE). The methodology is successfully applied to a real life benchmark: the Poland Electricity Load dataset.