Salvatore Vitabile

Efficient MLP digital implementation on FPGA

The efficiency and the accuracy of a digital feedforward neural networks must be optimized to obtain both high classification rate and minimum area on chip. In this paper an efficient MLP digital implementation. The key features of the hardware implementation are the virtual neuron based architecture and the use of the sinusoidal activation function for the hidden layer. The effectiveness of the proposed solutions has been evaluated developing different FPGA based neural prototypes for the high energy physics domain and the automatic road sign recognition domain. The use of the sinusoidal activation function decreases hardware resource employment of about 32% when compared with the standard sigmoid based neuron implementation. The virtual neuron implementation makes efficient the mapping of a neural network into hardware devices since it leads to a significant decreasing of concurrent memory access.

Image processing chain for digital still cameras based on the SIMPil architecture

The new generation of wireless devices herald the development of products for integrated portable image and video communication requiring to image and video applications high computing performance. Portable MultiMedia Supercomputers (PMMS), a new class of architectures, allow to combine high computational performance, needed by multimedia applications, and a big energy efficiency, needed by portable devices. Among PMMS, the SIMPil (SIMD processor pixel) architecture satisfies the above requirements, especially with video and digital images processing tasks. In this paper we, exploit the SIMPil computation and throughput efficiency to implement the whole image processing chain of a digital still camera device. The implemented chain covers the whole image pipeline: from the Bayer pattern image processing to the JPEG image compression. SIMPil performance has been evaluated using an instruction level simulator. To prove the effectiveness of the proposed approach, processing and compression results have been compared with the Texas Instruments Inc. TMS320C549 DSP one.

Efficient rapid prototyping of image and video processing algorithms

Image and video processing tasks are often confined for real-time execution on large size workstations or expensively custom designed hardware. The current availability of mature reconfigurable hardware, like field programmable gate arrays (FPGAs), coupled with the usage of hardware programming languages offers a good path for porting such applications on portable devices. This paper explores the rapid prototyping of a real-time road sign recognition system on a FPGA, using an algorithmic-like hardware programming language: the Handel-C language. We investigate the relationship between efficient Handel-C data, structures, constructs and the related high level C data, structures, constructs. Programming guidelines are proposed for the development of real-time image and video processing, starting from a better organized high level C code that can be then easily ported in Handel-C. Results are illustrated showing the effectiveness of employing Handel-C to turn an entirely software based system into a fully functional field deployable device.

An embedded biometric sensor for ubiquitous authentication

Communication networks and distributed technologies move people towards the era of ubiquitous computing. An ubiquitous environment needs many authentication sensors for users recognition, in order to provide a secure infrastructure for both user access to resources and services and information management. Today the security requirements must ensure secure and trusted user information to protect sensitive data resource access and they could be used for user traceability inside the platform. Conventional authentication systems, based on username and password, are in crisis since they are not able to guarantee a suitable security level for several applications. Biometric authentication systems represent a valid alternative to the conventional authentication systems providing a flexible e-infrastructure towards an integrated solution supporting the requirement for improved inter-organizational functionality. In this work the study and the implementation of a fingerprints-based embedded biometric system is proposed. Typical strategies implemented in Identity Management Systems could be useful to protect biometric information. The proposed sensor can be seen as a self-contained sensor: it performs the all elaboration steps on board, a necessary requisite to strengthen security, so that sensible data are securely managed and stored inside the sensor, without any data leaking out. The sensor has been prototyped via an FPGA-based platform achieving fast execution time and a good final throughput. Resources used, elaboration times of the sensor are reported. Finally, recognition rates of the proposed embedded biometric sensor have been evaluated considering three different databases: the FVC2002 reference database, the CSAI/Biometrika proprietary database, and the CSAI/Secugen proprietary database. The best achieved FAR and FRR indexes are respectively 1.07% and 8.33%, with an elaboration time of 183.32 ms and a working frequency of 22.5 MHz.

MLP Neural Network Implementation on a SIMD Architecture

An Automatic Road Sign Recognition System {A(RS)2} is aimed at detection and recognition of one or more road signs from realworld color images. The authors have proposed an A(RS)2 able to detect and extract sign regions from real world scenes on the basis of their color and shape features. Classification is then performed on extracted candidate regions using Multi-Layer Perceptron neural networks. Although system performances are good in terms of both sign detection and classification rates, the entire process requires a large computational time, so real-time applications are not allowed. In this paper we present the implementation of the neural layer on the Georgia Institute of Technology SIMD Pixel Processor. Experimental trials supporting the feasibility of real-time processing on this platform are also reported.

Fingerprint Image Enhancement Using Directional Morphological Filter

Fingerprint images quality enhancement is a topic phase to ensure good performance in an automatic fingerprint identification system (AFIS) based on minutiae matching. In this paper a new fingerprint enhancement algorithm based on morphological filter is introduced. The algorithm is based on three steps: directional decomposition, morphological filter and composition. The performance of the proposed approach has been evaluated on two sets of images: the first one is DB3 database from Fingerprint Verification Competition (FVC) and the second one is self collected using an optical scanner

A Concurrent Neural Classifier for HTML Documents Retrieval

A neural based multi-agent system for automatic HTML pages retrieval is presented. The system is based on the EαNet architecture, a neural network having good generalization capabilities and able to learn the activation function of its hidden units. The starting hypothesis is that the HTML pages are stored in networked repositories. The system goal is to retrieve documents satisfying a user query and belonging to a given class (i.e. documents containing the word “football” and talking about “Sports”). The system is composed by three interacting agents: the EαNet Neural Classifier Mobile Agent, the Query Agent, and the Locator Agent. The whole system was successfully implemented exploiting the Jade platform features and facilities. The preliminary experimental results show a good classification rate: in the best case a classification error of 9.98% is reached.

A neural multi-agent based system for smart HTML pages retrieval

A neural based multi-agent system for smart HTML page retrieval is presented. The system is based on the EalphaNet architecture, a neural network capable of learning the activation function of its hidden units and having good generalization capabilities. System goal is to retrieve documents satisfying a query and dealing with a specific topic. The system has been developed using the basic features supplied by the Jade platform for agent creation, coordination and control. The system is composed of four agents: the trainer agent, the neural classifier mobile agent, the interface agent, and the librarian agent. The sub-symbolic knowledge of the neural classifier mobile agent is automatically updated each time a new, previously not included, document topic is requested by the user. The neural classifier mobile agent also interacts with the librarian agent for retrieving the documents in the repositories and with the interface agent for user interaction. The proposed system is particularly useful for classifying documents stored in private networked document repositories that, for various reasons (i.e. privacy, security, and so on), cannot be indexed by an external search engine. The system is very efficient: the preliminary experimental results show that in the best case a classification error of 9.98% is obtained.

A vision agent for mobile robot navigation in time-variable environments

We present an architecture for mobile robot navigation based on Bayesian networks. The architecture allows a robot to plan the correct path inside an environment with dynamic obstacles. Interactions between the robot and the environment are based on a powerful vision agent. The results of simulations, showing the effectiveness of the approach, are described.

Fingerprint Registration Using Specialized Genetic Algorithms

One of the most common problem to realize a robust matching algorithm in an automated fingerprint identification system (AFIS) is the images registration. In this paper a fingerprints registration method based on a specialized genetic algorithm (GA) is proposed. A global transformation between two fingerprint images is performed using genetic data evolutions based on specialized mutation rate and solution refining. An AFIS including the above method has been developed and tested on two different fingerprint databases: NIST 4 ink-on-paper and self optical scanned. The obtained experimental results show that the proposed approach is comparable with literature systems working on medium quality fingerprints