A numerical study of two classes of neural network models is presented. The performance of Ising spin neural networks as content-addressable memories for the storage of bit patterns is analysed. By studying systems of increasing sizes, behaviour consistent with fintite-size scaling, characteristic of a first-order phase transition, is shown to be exhibited by the basins of attraction of the stored patterns in the Hopfield model. A local iterative learning algorithm is then developed for these models which is shown to achieve perfect storage of nominated patterns with near-optimal content-addressability. Similar scaling behaviour of the associated basins of attraction is observed. For both this learning algorithm and the Hopfield model, by extrapolating to the thermodynamic limit, estimates are obtained for the critical minimum overlap which an input pattern must have with a stored pattern in order to successfully retrieve it. The role of a neural network as a tool for optimising cost functions of binary valued variables is also studied. The particular application considered is that of restoring binary images which have become corrupted by noise. Image restorations are achieved by representing the array of pixel intensities as a network of analogue neurons. The performance of the network is shown to compare favourably with two other deterministic methods-a gradient descent on the same cost function and a majority-rule scheme-both in terms of restoring images and in terms of minimising the cost function. All of the computationally intensive simulations exploit the inherent parallelism in the models: both SIMD (the ICL DAP) and MIMD (the Meiko Computing Surface) machines are used.

Submitted by NATALIA CAROLINE LOPES TRAVASSOS null (nataliacaroline2006@yahoo.com.br) on 2017-02-15T14:56:05Z No. of bitstreams: 1 natalia_dissertação.pdf: 3422187 bytes, checksum: 73f8b94c641709f43b16401694318651 (MD5) / Approved for entry into archive by Juliano Benedito Ferreira (julianoferreira@reitoria.unesp.br) on 2017-02-21T17:52:48Z (GMT) No. of bitstreams: 1 travassos_ncl_me_ilha.pdf: 3422187 bytes, checksum: 73f8b94c641709f43b16401694318651 (MD5) / Made available in DSpace on 2017-02-21T17:52:48Z (GMT). No. of bitstreams: 1 travassos_ncl_me_ilha.pdf: 3422187 bytes, checksum: 73f8b94c641709f43b16401694318651 (MD5) Previous issue date: 2016-12-19 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Este trabalho apresenta um algoritmo de codificação para imagens comprimidas que representa cada pixel de uma imagem e suas coordenadas por um único valor. Para cada pixel e suas coordenadas, esse valor único é armazenado em um vetor que é usado na reconstrução da imagem sem que sua qualidade seja comprometida. O método proposto apresenta melhorias em relação a dois outros algoritmos propostos anteriormente, sendo que um deles já é uma melhoria do primeiro. O algoritmo apresentado neste trabalho difere dos outros dois algoritmos estudados na diminuição significativa do espaço necessário para armazenamento das imagens, na determinação de uma taxa de compressão exata e na redução do tempo de processamento de decodificação. Um outro avanço apresentado foi a compressão de imagens coloridas utilizando a ferramenta wavemenu em conjunto com o algoritmo que determina a taxa de compressão. / This work presents an encoding algorithm for compressed images that represents each pixel of an image and its coordinates by a single value. For each pixel and its coordinates, this unique value is stored in a vector that is used in the reconstruction of the image without its quality being compromised. The proposed method presents improvements in relation to two other algorithms previously proposed, one of which is already an improvement for the first one. The algorithm presented in this work differs from the other ones in the foollowing characteristcs: by the significant reduction of the space required for image storage, the determination of an exact compression rate and the reduction of the decoding processing time. Another advancement was the compression of colored images using the tool wavemenu in improvement with the algorithm that determines the compression ratio.

Orientador: Francisco Villarreal Alvarado / Resumo: Este trabalho apresenta um algoritmo de codificação para imagens comprimidas que representacada pixel de uma imagem e suas coordenadas por um único valor. Para cadapixel e suas coordenadas, esse valor único é armazenado em um vetor que é usado nareconstrução da imagem sem que sua qualidade seja comprometida. O método propostoapresenta melhorias em relação a dois outros algoritmos propostos anteriormente, sendoque um deles já é uma melhoria do primeiro. O algoritmo apresentado neste trabalho diferedos outros dois algoritmos estudados na diminuição significativa do espaço necessáriopara armazenamento das imagens, na determinação de uma taxa de compressão exata ena redução do tempo de processamento de decodificação. Um outro avanço apresentadofoi a compressão de imagens coloridas utilizando a ferramenta wavemenu em conjuntocom o algoritmo que determina a taxa de compressão. / Mestre

La sécurité est un problème important dans la transmission et le stockage de l'image, tandis que le cryptage est un moyen d'assurer la sécurité qui est requise dans de nombreuses applications (télévision par câble, la communication d'images militaires, systèmes d'imagerie confidentielle, etc.). Toutefois, à l'instar du message texte, les données images présentent des caractéristiques spéciales telles que la haute capacité, la redondance et la haute corrélation entre les pixels, et nécessite souvent une transmission et des traitements temps réel pour certaines applications. Construire un système rapide et efficace de cryptographie d'images suscite un intérêt considérable. C'est dans ce contexte qu’ont été menés ces travaux thèse qui portent sur l’élaboration d’un corrélateur optique en termes de cryptage/décryptage des données pour son implémentation dans un montage optique innovant. L’objectif de ces travaux est de réaliser un système optique de chiffrement sur la base d'exploitation de transformation optique et de générateurs chaotiques. L'idée originale des travaux consiste à exploiter la non-linéarité des systèmes chaotiques comme clés de chiffrement pour les systèmes optiques de chiffrement d'images multispectrales. Dans ces travaux de thèse, nous avons proposés et évalués plusieurs chiffrements d'images à base d’un système hyperchaotique et de transformées optiques (gyrator, Fourier, Baker , Arnold et Gerchberg- Saxton) à partir d’un processus de cryptage reposant sur une décomposition composants RVB et un encodage dans un flux dimensionnel d’images couleurs. L'originalité des solutions de chiffrement adoptée reposent sur l'exploitation de signaux réellement aléatoires à travers la mise en œuvre de générateurs hyperchaotiques pour la génération de données aléatoires sous forme images comme base de matrices de clés de chiffrement. En effet, ces générateurs présentent des propriétés et des caractéristiques fondamentales en termes de cryptage car il présente une non-linéarité, une imprédictibilité et une extrême sensibilité aux conditions initiales les rendant très intéressantes pour le développement de clés de chiffrement par flot. L’algorithme mis en œuvre permet d'extraire en temps réel les caractéristiques de texture dans les différentes bandes spectrales d'images en vue d’évaluer et de détecter les teneurs potentielles en information et dont les transmissions doivent être sécurisée via une transmission optique / Optical information security is one of the most important research directions in information science and technology, especially in the field of copyright protection, confidential information transmission/storage and military remote sensing. Since double random phase encoding technology (DRPE) was proposed, optical image encryption technology has become the main topic of optical information security and it has been developed and studied deeply. Optical encryption techniques offer the possibility of high-speed parallel processing of two dimension image data and hiding information in many different dimensions. In this context, much significant research and investigation on optical image encryption have been presented based on DRPE or further optical operation, such as digital holography, Fresnel transform, gyrator transform. Simultaneously, the encrypted image has been extended from single gray image to double image, color image and multi-image. However, the hyperspectral image, as a significant element in military and commercial remote sensing, has not been deeply researched in optical encryption area until now. This work extends the optical encryption technology from color image to hyperspectral image. For better comprehension of hyperspectral image encryption, this work begins with the introduction and analysis of the characteristics of hyperspectral cube. Subsequently, several kinds of encryption schemes for color image, including symmetric and asymmetric cryptosystem, are presented individually. Furthermore, the optical encryption algorithms for hyperspectral cube are designed for securing both the spatial and spectral information simultaneously. Some numerical simulations are given to validate the performance of the proposed encryption schemes. The corresponding attack experiment results demonstrate the capability and robustness of the approaches designed in this work. The research in this dissertation provides reference for the further practicality of hyperspectral image encryption