Alessio Ferone

Rough fuzzy set-based image compression

A new coding/decoding scheme based on the properties and operations of rough fuzzy sets is presented. By normalizing pixel values of an image, each pixel value can be interpreted as the degree of belonging of that pixel to the image foreground. The image is then subdivided into blocks which are partitioned and characterized by a pair of approximation sets. Coding uses a codebook, created with a quantization algorithm, to find the best approximating pair for each block, while decoding exploits specific properties of rough fuzzy sets to rebuild the blocks. The method, called by us rough fuzzy vector quantization (RFVQ) relies on the representation capabilities of the vector to be quantized and not on the quantization algorithm, to determine optimal codevectors. A comparison with other fuzzy-based coding/decoding schemes and with DCT and JPEG methods is performed by means of peak signal to noise ratio (PSNR) values. Results show that for low compression rates the proposed method performs well and, in some cases, the PSNR obtained with RFVQ is close to the JPEGs PSNR.

Neural Background Subtraction for Pan-Tilt-Zoom Cameras

We propose an extension of a neural-based background subtraction approach to moving object detection to the case of image sequences taken from pan-tilt-zoom (PTZ) cameras. The background model automatically adapts in a self-organizing way to changes in the scene background. Background variations arising in a usual stationary camera setting, such as those due to gradual illumination changes, to waving trees, or to shadows cast by moving objects, are accurately handled by the neural self-organizing background model originally proposed for this type of setting. Handling of variations due to the PTZ camera movement is ensured by a novel registration mechanism that allows the neural background model to automatically compensate the eventual ego-motion, estimated at each time instant. Experimental results on several real image sequences and comparisons with seven state-of-the-art methods demonstrate the accuracy of the proposed approach.

White Paper on Industrial Applications of Computer Vision and Pattern Recognition

The paper provides a summary of the contributions to the industrial session at ICIAP2013, describing a few practical applications of Video Analy- sis, in the Surveillance and Security field. The session has been organized to stimulate an open discussion within the scientific community of CVPR on new emerging research areas which deserve particular attention, and may contribute to the improvement of industrial applications in the near future.

OBJECT MOTION DETECTION AND TRACKING BY AN ARTIFICIAL INTELLIGENCE APPROACH

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.

Protein motifs retrieval by SS terns occurrences

This paper describes a new approach to the analysis of protein 3D structure based on the Secondary Structure (SS) representation. The focus is here on structural motif retrieval. The strategy is derived from the Generalized Hough Transform (GHT), but considering as structural primitive element, the triplet of SSs. The triplet identity is evaluated on the triangle having the vertices on the SS midpoints, and is represented by the three midpoints distances. The motif is characterized by the complete set of triplets, so the Reference Table (RT) has a tuple for each triplet. Tuples contain, beside the discriminant component (the three edge lengths), the mapping rule, i.e. the Reference Point (RP) location referred to the triplet. In the macromolecule to be analyzed, each possible triplet is searched in the RT and every match gives a contribution to a candidate location of the RP. Presence and location of the searched motif are certified by the collection of a number of contribution equal (obviously in absence of noise and ambiguities) to the RT cardinality (i.e. the number of motif triplets). The approach is tested on twenty proteins selected randomly from the PDB, but having a different number of SSs ranging from 14 to 46. The retrieval of all possible structural blocks composed by three, four and five SSs (very compact and completely distributed) have been conducted. The results show valuable performances for precision and computation time.

Search of protein structural blocks through secondary structure triplets

This paper presents an approach for protein motif retrieval founded on protein secondary structures (SSs) in 3D. This is a new way to analyze the protein 3D structure. In this approach, based on the Generalized Hough Transform (GHT), the primitives are the triangles defined by the midpoints of three SSs. The three distances between each SSs couple are used in searching and in the voting process. The barycenter of the motif is assigned as the Reference Point (RP). All motif triangles are compared with all possible triangles in the macromolecule. The lengths of triangle edges are used as selective parameters. For every correspondence a vote is given to the point which is figured out as motif barycenter with a special mapping rule and the point having most votes is determined as candidate RP. In this paper we made some experiments for retrieval of four- and five-SSs motif from the macromolecule. Experimental results showed that the RP is determined with precision and this new approach to retrieve the motif is simple to implement, computationally efficient and fast.

Searching Structural Blocks by SS Exhaustive Matching

This paper presents motif retrieval from a macromolecule or a protein by using structure comparison in 3D through an exhaustive matching analysis of secondary structures. The comparison is based on three parameters: midpoint distance (Md), axis distance (Ad) and angle (ϕ) related to a couple of SSs in 3D space. The barycenter of the motif is assigned as Reference Point (RP) and in order to find the RP related to every possible motif (instance) in the macromolecule a voting process is performed. The searched motif is compared with all possible instances having the same number of motif SSs in the macromolecule and gives a vote to the candidate barycenter for every correspondence. The point, which has the maximum number of votes, is determined as candidate RP. In this paper motifs composed by four and five secondary structures are searched. Experimental results show a good accuracy in determining the RP and hence in the retrieval of the searched motif.

Feature Discovery through Hierarchies of Rough Fuzzy Sets

Rough set theory and fuzzy logic are mathematical frameworks for granular computing forming a theoretical basis for the treatment of uncertainty in many real-world problems, including image and video analysis. The focus of rough set theory is on the ambiguity caused by limited discernibility of objects in the domain of discourse; granules are formed as objects and are drawn together by the limited discernibility among them. On the other hand, membership functions of fuzzy sets enables efficient handling of overlapping classes. The hybrid notion of rough fuzzy sets comes from the combination of these two models of uncertainty and helps to exploit, at the same time, properties like coarseness, by invoking rough sets, and vagueness, by considering fuzzy sets. We describe a model of the hybridization of rough and fuzzy sets, that allows for further refinements of rough fuzzy sets. This model offers viable and effective solutions to some problems in image analysis, e.g. image compression.

A real-time streaming server in the RTLinux environment using VideoLanClient

Accessing rich multimedia content through terminals and bandwidth-constrained networks is an important issue. Therefore, multimedia servers for delivering such data deserve much attention. A novel real-time streaming system for multimedia is presented, based on real-time operating systems. The VLC streaming server has been improved by employing a modified version of RTLinux that includes DMI and/or EDFI schedulers with the aim to build a complete streaming system for multimedia. Tests have been performed to measure one of the most important aspects of such systems, that is, jittering. The proposed system is able to achieve good performance both in simulated and real-world situations.

Neural Moving Object Detection by Pan-Tilt-Zoom Cameras

Automated video surveillance using video analysis and understanding technology has become an important research topic in the area of computer vision. Most cameras used in surveillance are fixed, allowing to only look at one specific view of the surveilled area. Recently, the progress in sensor technologies is leading to a growing dissemination of Pan-Tilt-Zoom (PTZ) cameras, that can dynamically modify their field of view. Since PTZ cameras are mainly used for object detection and tracking, it is important to extract moving object regions from images taken with this type of camera. However, this is a challenging task because of the dynamic background caused by camera motion.