International audience

Scene background initialization is the process by which a method tries to recover the background image of a video without foreground objects in it. Having a clear understanding about which approach is more robust and/or more suited to a given scenario is of great interest to many end users or practitioners. The aim of this paper is to provide an extensive survey of scene background initialization methods as well as a novel benchmarking framework. The proposed framework involves several evaluation metrics and state-of-the-art methods, as well as the largest video data set ever made for this purpose. The data set consists of several camera-captured videos that: 1) span categories focused on various background initialization challenges; 2) are obtained with different cameras of different lengths, frame rates, spatial resolutions, lighting conditions, and levels of compression; and 3) contain indoor and outdoor scenes. The wide variety of our data set prevents our analysis from favoring a certain family of background initialization methods over others. Our evaluation framework allows us to quantitatively identify solved and unsolved issues related to scene background initialization. We also identify scenarios for which state-of-the-art methods systematically fail.

Modern young people (“digital natives”) have grown in an era dominated by new technologies where communications are pushed to quite a real-time level, and pose no limits in establishing relationships with other people or communities. However, the speed of evolution does not allow young people to split consciously acceptable behaviors from potentially harmful ones and a new phenomenon known as cyber bullying is emerging with increasing evidence, attracting the attention of educators, and media. Cyber bullying is defined as “willful and repeated harm inflicted through the use of electronic devices” [1]. In this paper we propose a possible solution for automatic detection of bully traces over a social network, using techniques derived from NLP (Natural Language Processing) and machine learning. Specifically, we shall design a model inspired by Growing Hierarchical SOMs, able to cluster efficiently documents containing bully traces, built upon semantic and syntactic features of textual sentences. We fine-tuned our model to work with the social network Twitter, but we also tested the model against other social networks such as YouTube and Formspring. Finally, we report our results, showing that the proposed unsupervised approach could be effectively used with good performances in some scenarios.

In scene analysis, the availability of an initial background model that describes the scene without foreground objects is at the basis of many computer vision applications. Multi-modal models of the scene background are frequently adopted in the applications, where each mode tries to keep track of the multiple background modes observed along the sequence. In this work we specifically address the problem of extracting a single background image by a multi-modal model of the scene background, in order to compare it against a given ground truth image of the background. Experimental results are provided on the SBMnet dataset, based on an existing multi-model background model and different extraction criteria, and general conclusions are drawn.

Surveillance systems are widely spread in many public and private places of social life such as streets, offices, universities, hospitals and parking lots. The amount of data to be processed grows together with the need of skilled experts able to interpret it. For this reason, the scientific community put a great effort in conceiving approaches able to detect anomalies automatically. Whichever the systems that gather the data, be they heat sensors, cameras or gps trackers, anomalous event detection must rely on approaches able to promptly detect suspicious activities. To tackle this problem, we present an online point-wise approach to the analysis of spatio-temporal (trajectories) datasets in order to detect anomalous behaviors. We model trajectory dataset points as a Rough Set and use an outlier detection algorithm previously proposed. We perform a sliding temporal window scan to detect anomalies in real time. We demonstrate the robustness of this method using the CUHK crowd and the University of Udine trajectory datasets, which provide different combinations of trajectory scenarios: one group or multiple groups, one anomaly or multiple anomalies.

Deep convolutional neural networks (CNN) brought revolution without any doubt to various challenging tasks, mainly in computer vision. However, their model designing still requires attention to reduce number of learnable parameters, with no meaningful reduction in performance. In this paper we investigate to what extend CNN may take advantage of pyramid structure typical of biological neurons. A generalized statement over convolutional layers from input till fully connected layer is introduced that helps further in understanding and designing a successful deep network. It reduces ambiguity, number of parameters, and their size on disk without degrading overall accuracy. Performance are shown on state-of-the-art models for MNIST, Cifar-10, Cifar-100, and ImageNet-12 datasets. Despite more than 80% reduction in parameters for Caffe LENET, challenging results are obtained. Further, despite 10 - 20% reduction in training data along with 10 - 40% reduction in parameters for Alexnet model and its variations, competitive results are achieved when compared to similar well-engineered deeper architectures.

Graph-based semi-supervised learning methods play a key role in machine learning applications, particularly when no parametric information or other prior knowledge is available. Given a graph whose nodes represent the points and the weighted edges the relations between them, the goal is to predict the values of all unlabeled nodes exploiting the information provided by both label and unlabeled nodes. In this paper, we propose a novel graph-based approach for semi-supervised binary classification. The algorithm extends the Fisher Subspace estimation approaches by adopting a kernel graph covariance measure. This similarity measure defines a relation between nodes generalizing both the shortest path and the commute time distance. This quantity is called the sum-over-paths covariance. Experiments on synthetic and real-world datasets highlight that the proposed algorithm achieves better results with respect to those obtained by state-of-the-art competitors.

Nowadays, the high availability of data gathered from wireless sensor networks and telecommunication systems has drawn the attention of researchers on the problem of extracting knowledge from spatiotemporal data. Detecting outliers which are grossly different from or inconsistent with the remaining spatiotemporal data set is a major challenge in real-world knowledge discovery and data mining applications. In this paper, we deal with the outlier detection problem in spatiotemporal data and describe a rough set approach that finds the top outliers in an unlabeled spatiotemporal data set. The proposed method, called Rough Outlier Set Extraction (ROSE), relies on a rough set theoretic representation of the outlier set using the rough set approximations, i.e., lower and upper approximations. We have also introduced a new set, named Kernel Set, that is a subset of the original data set, which is able to describe the original data set both in terms of data structure and of obtained results. Experimental results on real-world data sets demonstrate the superiority of ROSE, both in terms of some quantitative indices and outliers detected, over those obtained by various rough fuzzy clustering algorithms and by the state-of-the-art outlier detection methods. It is also demonstrated that the kernel set is able to detect the same outliers set but with less computational time.

Granular information processing is one of the human-inspired problem-solving aspects of natural computing, as information abstraction is inherent in human thinking and reasoning processes, and plays an essential role in human cognition. Among the different facets of natural computing fuzzy sets, rough sets and their hybridization are well accepted paradigms that are based on the construction, representation and interpretation of granules, as well as the utilization of granules for problem solving. These tools are also known as primary constituents of soft computing whose objective is to provide flexible information processing capability for handling real-life ambiguous situations. They have been successfully employed in various image processing tasks, including image segmentation, enhancement and classification, both individually or in combination with other computing techniques. The reason of such success is rooted to the fact that they provide powerful tools to describe uncertainty, naturally embedded in images, which can be exploited in various image processing tasks.

The automatic detection of objects that are abandoned or removed in a video scene is an interesting area of computer vision, with key applications in video surveillance. Forgotten or stolen luggage in train and airport stations and irregularly parked vehicles are examples that concern significant issues, such as the fight against terrorism and crime, and public safety. Both issues involve the basic task of detecting static regions in the scene. We address this problem by introducing a model-based framework to segment static foreground objects against moving foreground objects in single view sequences taken from stationary cameras. An image sequence model, obtained by learning in a self-organizing neural network image sequence variations, seen as trajectories of pixels in time, is adopted within the model-based framework. Experimental results on real video sequences and comparisons with existing approaches show the accuracy of the proposed stopped object detection approach.

Standards are essential to guarantee reusability and interoperability of e-learning content. Course designers should use standards in each step of the production of a learning unit, including documentation and drafting. While UML diagrams have already been proposed in web-based e-learning modeling, the navigational (dynamic) and structural (static) aspects have been kept separated. We propose the use of a UML profile that, together with course structure, semantically expresses different possible choices in user learning path. A real world application to a University course is presented.

Personalized Web page recommendation is strongly limited by the nature of web logs, the intrinsic complexity of the problem and the tight efficiency requirements. When tackled by traditional Web usage mining techniques, due to the presence of an huge number of meaningful clusters and profiles for visitors of a typical highly rated Website, the model-based or distance-based methods tend to make too strong and simplistic assumptions or, conversely, to become excessively complex and slow. In this paper, a heuristic ¿majority intelligence¿ strategy is designed, that easily adapts to changing navigational patterns, without the costly need to explicitly individuate them before navigation. The proposed approach mimics human behavior in an unknown environment in presence of many individuals acting in parallel and is able to predict with good accuracy and in real time the next page category visited by a user. The method has been tested on real data coming from users who visited a popular Website of generic content. Average accuracy on test sets is good on a 17 class problem and, most remarkably, it remains stable as the Web navigation goes on.

The aim of this paper is to propose an artificial intelligence based approach to moving object detection and tracking. Specifically, we adopt an approach to moving object detection based on self organization through artificial neural networks. Such approach allows to handle scenes containing moving backgrounds and gradual illumination variations, and achieves robust detection for different types of videos taken with stationary cameras. Moreover, for object tracking we propose a suitable conjunction between Kalman filtering, properly instanced for the problem at hand, and a matching model belonging to the class of Multiple Hypothesis Testing. To assess the validity of our approach, we experimented both proposed moving object detection and object tracking over different color video sequences that represent typical situations critical for video surveillance systems.

Detection of moving objects in video streams is the first relevant step of information extraction in many computer vision applications. Aside from the intrinsic usefulness of being able to segment video streams into moving and background components, detecting moving objects provides a focus of attention for recognition, classification, and activity analysis, making these later steps more efficient. We propose an approach based on self organization through artificial neural networks, widely applied in human image processing systems and more generally in cognitive science. The proposed approach can handle scenes containing moving backgrounds, gradual illumination variations and camouflage, has no bootstrapping limitations, can include into the background model shadows cast by moving objects, and achieves robust detection for different types of videos taken with stationary cameras. We compare our method with other modeling techniques and report experimental results, both in terms of detection accuracy and in terms of processing speed, for color video sequences that represent typical situations critical for video surveillance systems.

Image descriptions aimed at the realization of content-based image retrieval (CBIR) should include the vagueness of both data representations and user queries. Here we show how multiscale textural gradient can be used as an efficient visual cue for image description. This feature has been already efficiently used in problems of image segmentation and texture separation. Our main idea is based on the assumption that, for image description, shape and textures should be considered together within a unified model. We report an efficient image description algorithm where the multiscale analysis is modeled by a differential morphological filter. Experiments with large image databases and comparisons with classical methods are reported.

A fundamental step of microarray image analysis is the detection of the grid structure for the accurate localization of each spot, representing the state of a given gene in a particular experimental condition. This step is known as gridding or microarray addressing. Most of the available microarray gridding approaches require human intervention; for example, to specify landmarks, some points in the spot grid, or even to precisely locate individual spots. Automating this part of the process can allow high throughput analysis (Yang, Y, et al, 2002). This paper is aimed towards at the development fully automated procedures for the problem of automatic microarray gridding. Indeed, many of the automatic gridding approaches are based on two phases, the first aimed at the generation of an hypothesis consisting into a regular interpolating grid, whereas the second performs an adaptation of the hypothesis. Here we show that the first step can efficiently be accomplished by using the Radon transform, whereas the second step could be modeled by an iterative posterior maximization procedure (Antoniol, G and Ceccarelli, M, 2004).

Blotchy noise is one of the most common and visually annoying noises noticed in digitized motion picture films. This work investigates a three-stage blotchy noise reduction scheme combining efforts of temporal median filtering, ICA unsupervised learning and deflation after motion/compensation estimation. Implementation of each module of the scheme is discussed; finally performance of the scheme is tested and compared with other methods over real short sequences and results are discussed.

Fuzzy neural systems have been a subject of great interest in the last few years, due to their abilities to facilitate the exchange of information between symbolic and subsymbolic domains. However, the models in the literature are not able to deal with structured organization of information, that is typically required by symbolic processing. In many application domains, the patterns are not only structured, but a fuzziness degree is attached to each subsymbolic pattern primitive. The purpose of this paper is to show how recursive neural networks, properly conceived for dealing with structured information, can represent nondeterministic fuzzy frontier-to-root tree automata. Whereas available prior knowledge expressed in terms of fuzzy state transition rules are injected into a recursive network, unknown rules are supposed to be filled in by data-driven learning. We also prove the stability of the encoding algorithm, extending previous results on the injection of fuzzy finite-state dynamics in high-order recurrent networks.

In recent years, there have been several proposals for the realization of models inspired to biological solutions for pattern recognition. In this work we propose a new approach, based on a hierarchical modular structure, to realize a system capable to learn by examples and recognize objects in digital images. The adopted techniques are based on multiresolution image analysis and neural networks. Performance on two different data sets and experimental timings on a single instruction multiple data (SIMD) machine are also reported.

The paper reports a correlation-based method for the detection of circular objects which is capable of overcoming well-known problems arising from the use of gradient-based voting schemes. Specifically, the method is: (a) capable of detecting circular objects on the basis of both magnitude and direction of the image gradient; and (b) dealing with three-dimensional spherical objects by considering shadows depending on the direction of light. Experimental results on the accuracy of the method and comparisons with the Hough transform and the Hausdorff matching are reported.

Reports a texture separation algorithm to solve the problem of unsupervised boundary localization in textured images. The proposed algorithm is mainly characterized by the extraction of textural density gradients by a nonlinear multiple scale-space analysis of the image. Texture boundaries are extracted by segmenting the images resulting from a multiscale fuzzy gradient operation applied to detail images. The segmentation stage consists of a parallel hierarchical clustering algorithm, aimed at the minimization of a global cost functional taking into account region homogeneity and segmentation quality. Experiments and comparisons on Brodatz textures are reported.

A new parallel thinning algorithm with two subcycles is proposed and compared with other parallel thinning algorithms in terms of 8-connectedness degree, erosion, stability under pattern rotation, and boundary noise sensitivity. Computational issues are also reported based on the implementation of the thinning algorithm on the SIMD machines CM-200 and MasPar MPP-12000.

Abstract The paper reports two techniques for parallelizing on a MIMD multicomputer a class of learning algorithms (competitive learning) for artificial neural networks widely used in pattern recognition and understanding. The first technique presented, following the divide et impera strategy, achieves O(n/p + logP) time for n neurons and P processors interconnected as a tree. A modification of the algorithm allows the application of a systolic technique with the processors interconnected as a ring; this technique has the advantage that the communication time does not depend on the number of processors. The two techniques are also compared on the basis of predicted and measured performance on a transputer‐based MIMD machine. As the number of processors grows the advantage of the systolic approach increases. On the contrary, the divide et impera approach is more advantageous in the retrieving phase.