Giuseppe Mastronardi

A face recognition system based on Pseudo 2D HMM applied to neural network coefficients

Face recognition from an image or video sequences is emerging as an active research area with numerous commercial and law enforcement applications. In this paper different Pseudo 2-dimension Hidden Markov Models (HMMs) are introduced for a face recognition showing performances reasonably fast for binary images. The proposed P2-D HMMs are made up of five levels of states, one for each significant facial region in which the input frontal images are sequenced: forehead, eyes, nose, mouth and chin. Each of P2-D HMMs has been trained by coefficients of an artificial neural network used to compress a bitmap image in order to represent it with a number of coefficients that is smaller than the total number of pixels. All the P2-D HMMs, applied to the input set consisting of the Olivetti Research Laboratory face database combined to others photos, have achieved good rates of recognition and, in particular, the structure 3-6-6-6-3 has achieved a rate of recognition equal to 100%.

Extending Hough Transform to a Points' Cloud for 3D-Face Nose-Tip Detection

This paper describes an extension of Generalized Hough Transform (GHT) to 3D point cloud for biometric applications. We focus on the possibility to applying this new GHT on a dataset representing point clouds of 3DFaces in order to obtain a nose-tip detection system.

Retinal Fundus Biometric Analysis for Personal Identifications

In this paper a biometric system for personal identification, realized through the manipulation of retinal fundus images and the detection of its bifurcation points, is described. In the image pre-processing step, a strong contrast exaltation between blood vessels and the background in retinal image is carried out; then blood vessels are extracted and next the vasculature bifurcation and crossover points are identified within squared shaped regions used to window the image. Finally the features sets are compared with a pattern recognition algorithm and a novel formulation is introduced to evaluate a similarity score and to obtain the personal identification.

Novel Data Mining Techniques in aCGH based Breast Cancer Subtypes Profiling: the Biological Perspective

In this paper we present a comparative study among well established data mining algorithm (namely J48 and naive Bayes tree) and novel machine learning paradigms like ant miner and gene expression programming. The aim of this study was to discover significant rules discriminating ER+ and ER-cases of breast cancer. We compared both statistical accuracy and biological validity of the results using common statistical methods and gene ontology. Some worth noting characteristics of these systems have been observed and analysed even giving some possible interpretations of findings. With this study we tried to show how intelligent systems can be employed in the design of experimental pipeline in disease processes investigation and how deriving high-throughput results can be validated using new computational tools. Results returned by this approach seem to encourage new efforts in this field

Steganography effects in various formats of images. A preliminary study

In this paper, the effects of steganography in different image formats (BMP, GIF, JPEG and DWT coded) are studied. With respect to these formats, we try to give an answer to the following questions. (1) How many bits of noise (i.e. the textual secret message) can be injected without perceptually deteriorating the quality of the image? (2) How and where should one inject these bits in order to achieve the best trade-off in terms of the length of the textual message and the preserved quality of the image?.

Retinal vessel extraction by a combined neural network-wavelet enhancement method

This paper presents a combined approach to automatic extraction of blood vessels in retinal images. The proposed procedure is composed of two phases: a wavelet transform-based preprocessing phase and a NN-based one. Several neural net topologies and training algorithms are considered with the aim of selecting an effective combined method. Human retinal fundus images, derived from the publicly available ophthalmic database DRIVE, are processed to detect retinal vessels. The approach is tested by considering performances in terms of sensitivity and specificity values obtained from vessel classification. The quality of vessel identifications is evaluated on obtained image by computing both sensitivity values and specificity ones and by relating them in ROC curves. A comparison of performances by ROC curve areas for various methods is reported.

Face Detection by Means of Skin Detection

In this article we present a novel approach to detect face in color images. Many researchers concentrated on this problem and the literature about this subject is extremely wide. We thought to decompose the overall problem into two intuitive sub-problems: the research of pixels have skin color in the original image and the analysis of pixels portions resulting, by means of a neural classifier. The achieved results show the robustness of presented approach.

Pseudo 2D Hidden Markov Models for Face Recognition Using Neural Network Coefficients

Face recognition is the preferred mode of identity recognition by humans from an image or video sequence: it is natural, robust and unintrusive. This work presents different pseudo 2D HMM structures for a face recognition showing performances reasonably fast for binary image. The proposed P2-D HMMs are made up of five levels of states, one for each region of interest (Rol) in which the input frontal images are sequenced: forehead, eyes, nose, mouth and chin. Each of P2-D HMMs has been trained by coefficients of an artificial neural network used to compress a bitmap image in order to represent it with a number of coefficients that is smaller than the total number of pixels. All the P2-D HMMs, applied to the validation set consisting of the Olivetti Research Laboratory (ORL) face database, have achieved good rates of recognition compared to other methods proposed in the literature and, in particular, the structure 3-6-6-6-3 has achieved a rate of recognition equal to 100%.

Intrusion Detection Using Neural Networks: A Grid Computing Based Data Mining Approach

Scientific disciplines such as life sciences as well as security and business fields depend on Knowledge Discovery because of the increasing amount of data being collected and for the complex analyses that need to be performed on them. New techniques, such as parallel, distributed, and grid-based data mining, are often able to overcome some of the characteristics of current data sources such as their large scale, high dimensionality, heterogeneity, and distributed nature. In several of these data mining applications, neural networks can be successfully applied. Moreover, an approach using neural networks seems to be one of the most promising methods for intrusion detection in a computer system or network security today. In this paper we describe a grid computing data mining approach for an intrusion detection application based on neural networks. Detection is carried out through the analyses of internet traffic generated by users in a network computer system.

Image processing framework for virtual colonoscopy

This paper describes a complete image processing framework for Virtual Colonscopy. The developed algorithms cover the entire process that allows a virtual navigation inside the colon lumen, starting from a dataset of axial CT slices. The implemented modules are: electronic colon cleansing, lumen segmentation, skeletonization, rendering and navigation. In particular for the centerline problem two different techniques are proposed and evaluated.