Early approaches based on local features work on image blocks. In [3] color and texture features are extracted from image blocks to train a statistical model; this model takes into account spatial relations among blocks and across image resolutions. Then several region-based approaches have been proposed, which require segmentation of images into relevant regions. E.g. in [4], an algorithm for learning region prototypes is proposed as well as a classification of regions based on Support Vector Machines (SVMs). More recent techniques have suc- cessfully used bags-of-features, which collect local features into variable length vectors. In [5], the bags-of-features representing an image are spatial pyramid aggregating statistics of local features (e.g. “SIFT” **descriptors**); this approach takes into account approximate global geometric correspondences between local features. Other methods using bags-of-features are based on explicitly modeling the distribution of these vector sets. In fact, measuring the similarity between the bags-of-features’ distributions is the main difficulty for this kind of catego- rization methods. For example, Gaussian Mixture Models (GMMs) have been used to model the distribution of bags of low-level features [6]. This approach requires both to estimate the model parameters and to compute a similarity measure to match the distributions.

En savoir plus
Recognizing spatio-temporal similarities between video sequences is a difficult task, since the objects undergo var- ious transformations or may be occulted through the se- quence. Our objective in this paper is to provide a frame- work that will enable 1) to answer to different search task on video databases (e.g. find videos with similar motion or videos containing a similar object) and 2) to provide coher- ent answers with the various data formats that are available to the user through a heterogeneous network (i.e. give simi- lar answer whether the user is sending its query from a PDA or his desktop computer). To do so, we define two kinds of **descriptors**, 1) global visual **descriptors** - also called spatial **descriptors** - that capture the visual content of a scene and 2) temporal or motion **descriptors** that capture the trajectories of objects in the videos. Both kinds of **descriptors** are patch **descriptors** that exploit respectively the spatial and tempo- ral coherence present in the video. The sets of **descriptors** are compared statistically by a dissimilarity measure so that loose transformations of the video are not penalized.

En savoir plus
We applied this framework to image retrieval. The proposed approach takes advantage of the properties of its global **multiscale** **descriptors**. In particular, it is robust to JPEG2000 compression (i.e. it matches the visual similarity be- tween images with different amounts of blur or compression noise). Retrieval experiments were conducted on two publicly available datasets of real world images (Nister Recognition Benchmark and the COREL database) to evaluate the average performances of the method. In particular, the Nister dataset was used to benchmark the robustness to several geometric image deformations, such as change of viewpoint, rotation and zoom. Our results showed the reliability of the SMP approach with respect to these deformations. In addition, although our method is new, its performances tested on two databases are very close to those of several established retrieval methods: a reference retrieval method based on (local) SIFT **descriptors** and two versions of a fuzzy, segmentation-based UFM approach: UFM and CLUE . This indicates that the SMP approach adapts to quite different retrieval tasks, from the object level (on the Nister database) to the level of general categories (on the COREL database). Finally, our **Sparse** **Multiscale** **Patches** approach follows the same **multiscale** philosophy as the new compression standard SVC [38], presuming nearly straightforward use of low- level bitstream components for retrieval purposes. This framework can also be used for other image processing tasks such as tracking or denoising.

En savoir plus
HYBRID HIGH-ORDER METHODS
TH ´ EOPHILE CHAUMONT-FRELET † , ALEXANDRE ERN ] , SIMON LEMAIRE [ ,
AND FR ´ ED ´ ERIC VALENTIN ‡
Abstract. We establish the equivalence between the **Multiscale** Hybrid-Mixed (MHM) and the **Multiscale** Hybrid High-Order (MsHHO) methods for a variable diffusion problem with piecewise polynomial source terms. Under the idealized assumption that the local problems defining the **multiscale** basis functions are exactly solved, we prove that the equivalence holds for general polytopal (coarse) meshes and arbitrary approximation orders. Also, we leverage the interchange of properties to perform a unified convergence analysis, as well as to improve on both methods.

En savoir plus
4. Coupling
In this section we review a well known technique that can be then adapted to prove that some operation produce **sparse** sets.
Let A and B be two finite sets and let R ⊂ A × B be a binary relation; we write aRb if ha, bi ∈ R. Consider two random variables α and β that range over A and B. defined on unrelated probability spaces. We say that αRβ (abusing slightly the notation) if there exist random variables α ′ and β ′ defined on some common probability space M such that

Abstract
The conservation of the coastal marine environment requires the possession of information that enables the global quality of the environment to be evaluated reliably and relatively quickly. The use of biological indicators is often an appropriate method. Seagrasses in general, and Posidonia oceanica meadows in particular, are considered to be appropriate for biomonitoring because of their wide distribution, reasonable size, sedentary habit, easy collection and abundance and sensitivity to modifications of littoral zone. Reasoned management, on the scale of the whole Mediterranean basin, requires standardized methods of study, to be applied by both researchers and administrators, enabling comparable results to be obtained. This paper synthesises the existing methods applied to monitor P. oceanica meadows, identifies the most suitable techniques and suggests future research directions. From the results of a questionnaire, distributed to all the identified laboratories working on this topic, a list of the most commonly used **descriptors** was drawn up, together with the related research techniques (e.g. standardization, interest and limits, valuation of the results). It seems that the techniques used to study meadows are rather similar, but rarely identical, even though the various teams often refer to previously published works. This paper shows the interest of a practical guide that describes, in a standardized way, the most useful techniques enabling P. oceanica meadows to be used as an environmental descriptor. Indeed, it constitutes the first stage in the process.

En savoir plus
I. I NTRODUCTION
In order to make the multimedia data searchable by its con- tent, various methods of mapping the multimedia content into high-dimensional spaces have been introduced for images [4] and audio [7]. Since, like all high dimensional data suffer from the curse of dimensionality, we would like to analyze such data to understand its nature and to give other researchers a base ground for further work, e.g., indexing. In [2] was proven that the complexity of searching the data grows exponentially with the dimensionality of data thus it is important to be able to set the tradeoff between fine grained information as high- dimensional feature vectors and good searchability of the data. Therefore, in this paper we present a comparative study of the properties of multimedia datasets representing visual and audio **descriptors** acquired from the public domain content provided by EWA 1 . The data is investigated in terms of pair- wise distance distribution and of the estimation of the intrinsic dimensionality. Because we focus on multimedia in general, we incorporate in our study both visual and audio data. Using the same methodology and criteria and by comparing the results, we would like to depict the different characteristics of these two types of multimedia considering also the datasets where the characteristics is known.

En savoir plus
Remark. In contrast to the bifurcation from the disc where we get a collection of m folds, the V-states of Theorem 1 are in general one or two-folds. For m even we can show from the proof that the V-states are symmetric with respect to the origin.
Now we shall sketch the proof of Theorem 1 which is mainly based upon the bifurcation theory via Crandall-Rabinowitz theorem. We shall look for a parametrization of the boundary ∂D of the rotating **patches** as a small perturbation of a given ellipse. This parametrization takes the form Φ : T → ∂D, with T is the unit circle and

The fruit fly Drosophila melanogaster has emerged as a model organism for research on social interactions. Although recent studies have described how individuals interact on foods for nutrition and reproduction, the complex dynamics by which groups initially develop and disperse have received little attention. Here we investigated the dynamics of collective foraging decisions by D. melanogaster and their variation with group size and composition. Groups of adults and larvae facing a choice between two identical, nutritionally balanced food **patches** distributed themselves asymmetrically, thereby exploiting one patch more than the other. The speed of the collective decisions increased with group size, as a result of flies joining foods faster. However, smaller groups exhibited more pronounced distribution asymmetries than larger ones. Using computer simulations, we show how these non-linear phenomena can emerge from social attraction towards occupied food **patches**, whose effects add up or compete depending on group size. Our results open new opportunities for exploring complex dynamics of nutrient selection in simple and genetically tractable groups. KEY WORDS: Aggregation, Drosophila melanogaster, Collective behavior, Foraging, Fruit flies, Individual-based model, Social attraction

En savoir plus
Electrochemical incompatibility of **patches** in reinforced concrete
Gu, P.; Beaudoin, J. J.; Tumidajski, P. J.; Mailvaganam, N. P.
https://publications-cnrc.canada.ca/fra/droits
L’accès à ce site Web et l’utilisation de son contenu sont assujettis aux conditions présentées dans le site LISEZ CES CONDITIONS ATTENTIVEMENT AVANT D’UTILISER CE SITE WEB.

To cite this version : Durham, William M. and Climent, Eric and Barry, Michael and De Lillo,
Filipo and Boffetta, Guido and Cencini, Massimo and Stocker, Roman Turbulence drives
microscale **patches** of motile phytoplankton. (2013) Nature Communications, vol. 4 . pp. 1-7.
ISSN 2041-1723

B. Fisher vector
By counting the number of occurrences of visual words, BOW encodes the 0-order statistics of the distribution of **descriptors**. The Fisher vector extends the BOW by encoding high-order statistics (first and, optionally, second order). This description vector is the gradient of the sample’s likelihood with respect to the parameters of this distribution, scaled by the inverse square root of the Fisher information matrix. As a result, it gives the direction, in parameter space, into which the learned distribution should be modified to better fit the observed data. In other terms, FV describes how the set of de- scriptors deviates from an average distribution of **descriptors**, modeled by a parametric generative model. It has been shown [20] that discriminative classifiers can be learned in this new representation space. In our image search framework, the FV is seen as a method to capture the information conveyed by a set of **descriptors** into a fixed-length signature.

En savoir plus
Abstract: In this research report we propose a novel approach to build interest points and **descriptors** which are invariant to a subclass of affine transformations. Scale and rotation invariant interest points are usually obtained via the linear scale-space representation, and it has been shown that affine invariance can be achieved by warping the smoothing kernel to match the local image structure. Our approach is instead based on the so-called Affine Morphological Scale-Space, a non-linear filtering which has been proved to be the natural equivalent of the classic linear scale-space when affine invariance is required. Simple local image **descriptors** are then derived from the extracted interest points. We demonstrate the proposed approach by robust matching experiments.

En savoir plus
The basic idea is to compute features inspired from the visual system model and specially from the main characteristics of the retina processing. Such descriptor is well adapted in the case of our cells images since the most discriminative visual feature between categories is the luminance contrast in sub-cellular regions. Thus, we define cell **descriptors** based on the local contrast in the cell that we call BIF (Bio-Inspired Features) [1]. The local contrast is obtained by a filtering with Dif- ferences of Gaussians (DoGs) centered at the origin. So that the contrast C Im for each position (x, y) and

En savoir plus
[ 20 ] Interestingly, based on longitudinal cross‐correlation,
Wardinski and Korte [2008] found in some periods opposing trends in the northern/southern hemispheres. Dumberry and Bloxham [2006] also argued that the equatorial symmetry is broken on millennial timescales. This is in contradiction to equatorially symmetric columnar flow that is thought to prevail in Earth’s core on shorter timescales due to the expected dominance of rotational effects [Busse, 1975; Olson et al., 1999; Jault, 2008; Pais and Jault, 2008]. A pair of prominent intense flux **patches** in both hemispheres near 90°W (Figure 5) is possibly the surface expression of columnar flow. To test this hypothesis, we focus on periods in which both **patches** are simultaneously present in Figure 5. Between −800 to −600 and between −50 to 100 the northern patch moves eastward, whereas the southern generally moves westward. Between 1250 to 1850 the northern patch drifts westward, whereas the southern alternates between eastward and westward motions. Overall, these results do not support equatorial symmetry in the kinematics of intense flux **patches**. It is worth stressing, however, that the kinematics of the high‐latitude intense flux **patches** does not necessarily reflect core flow, but could alternatively result from wave Figure 6. Polar views of **patches** trajectories corresponding

En savoir plus
We propose a generic scheme for the 3D reconstruction of planar **patches** from a single stereo pair. It does not require any additional localized information. The method assumes that region segmentation has been performed on both images, as well as the matching between 2D primitives, with methods such as described in [AMG93] and [RG91].

5 Conclusion
Our modulation strategy integrates the dominant orientation directly in the cod- ing stage. It is inspired by and builds upon recent works on explicit feature maps and kernel **descriptors**. Thanks to a generic formulation provided by match ker- nels, it is compatible with coding strategies such as Fisher vector or VLAD. Our experiments demonstrate that it gives a consistent gain compared to the original coding in all cases, even after dimensionality reduction. Interestingly, it is also very effective with a simple monomial kernel, offering competitive performance for image search with a coding stage not requiring any quantization.

En savoir plus
Keywords: illuminance, interior lighting, lighting design & specifications, luminance, uniformity
1.0 INTRODUCTION
Since the workshop held at the 23rd Session of the CIE in 1995, lighting quality has been established as one of the fundamental problems in lighting design, research, and education. The workshop at the 24th Session addressed a topic identified by delegates at the First CIE Symposium on Lighting Quality, held in 1998 [1]: the need to refine our definitions of luminous conditions to develop useful **descriptors** of the luminous field. Agreement about these photometric **descriptors** is a necessary first step to the development of research programs to advance understanding of the effects of the luminous environment on human behaviour, mood, and health, and to the development of recommendations for good-quality lighting design.

En savoir plus
Both the number of categories and the number of examples per category influence the performance. To investigate the influence of decreasing only the number of examples, we repeated the experi- ments on subsets of ucid5–ucid3, using less examples per category. These results are shown in Fig. 3. As expected, the performance of both **descriptors** declines both for increasing the number of cat- egories, and decreasing the number of examples while keeping the category number constant. However, the rate of precision decline w. r. t. number of examples per category looks moderately lower for the LPS than for SIFT, indicating that using less examples with LPS

En savoir plus
functions that can be substituted by different descriptor terms, such as polar description, B-spline and Fourier **descriptors**.
As we did in [6], the evolution of the curve is modeled according to the Lagrangian mechanics formalism by consid- ering q as the generalized coordinates of the system. The Euler-Lagrange equation of each component q i is defined by: