Catálogo de publicaciones - libros

Image scramble is one of the key technologies in the digital watermarks. In the absence of standardization and specific requirements in Watermark procedures, there are many methods for image scramble. In order to improve robustness, secrecy and exclusion of scrambled image, a method of neural network image scramble for wavelet watermark is proposed in this paper. Because the nonlinear mapping from the original image to its scrambled image is the key for image scramble, the nonlinear feature of neural network can be used to get the scrambled image. Our method embeds watermark into the wavelet descriptors. Watermark scrambled image generated by neural network can be successfully detected even after rotation, translation, scaling. And watermarks of our scheme are good at defending many kind watermark attacks. The experimental results demonstrate that our watermark algorithm is useful and practical.

The technological advances in the production and storage of fishery products have exceeded the development of effective market demand over the past one-decade. As a result, participants within the fishery industry have frequently found themselves facing increased variable and declining prices negatively affected the fishery industry and need to be pro-active instead of reactive to market changes. In this paper, a hybrid model is described, which integrate the Wavelet Neural Network with Genetic Algorithm and can predict the short-term aquatic products price. Then the theory framework and algorithms of the model are discussed. Then an empirical example is described. It shows that the proposed model can predict the short-term aquatic product price with the scale of one day, one week and ten days and the precision of prediction is not the decline trend when the forecasting scale is extended.

This paper investigates the effectiveness of a parallelized approach to VQ based image compression. In particular, we consider an image compression method using multiple VQs. The method, called MVQ, generates multiple independent codebooks to compress an image by using a neural network algorithm. In the image restoration, MVQ restores low quality images from the multiple codebooks, and then combines the low quality ones into a high quality one. Further, we present an effective coding scheme for codebook indexes to overcome the in-efficiency of MVQ in compression rate. Our simulation results show that the MVQ method outperforms a conventional single-VQ method when the compression rate is smaller than some values.

A mixture multiscale autoregressive moving average (ARMA) network is proposed for unsupervised segmentation of synthetic aperture radar (SAR) image. The network combines the multiscale analysis (MA) method and the feedforward artificial neural network (FANN), thus maintains some of the characteristics of the MA method and the FANN respectively. A corresponding learning algorithm is derived based on the Akaike’s information criterion (AIC) and genetic algorithm (GA). Experimental results on SAR images are shown to validate the presented network and learning algorithm.

In order to estimate the change of brain activity under general anesthesia, the Lempel-Ziv complexity (C(n)) of electroencephalogram (EEG) of SD rat was studied in this paper. The C(n)s of EEG from different channels under different depth of anesthesia were measured and the relationship between C(n) and the depth of anesthesia (DOA) was analyzed. The result shows that the C(n) variations of EEG of different channels with DOA are similar, and predicates that the activities of every part of brain change similarly. Therefore, it is enough to detect DOA by only one channel EEG. The C(n) of EEG will decrease while the depth of anesthesia increasing and vice versa. Two thresholds of C(n) are defined, one distinguishes awake and light anesthesia state, the other distinguishes light and deep anesthesia state. Besides EEG complexity analysis, the complexity variations of four rhythms of EEG (delta, theta, alpha and beta) are also analyzed. The study shows the dynamic change of complexity of delta rhythm leads to that of EEG, so the delta rhythm is the dominant rhythm during anesthesia for rat.

Blind source separation of post-nonlinear mixtures is discussed. The demixing system of the post-nonlinear mixtures is modeled using a multi-input multi-output wavelet neural network whose parameters can be determined under the criterion of independence of its outputs. A criterion of independence based on higher order moments is used to measure the statistical dependence of the outputs of the demixing system, and the particle swarm optimization technique is utilized to minimized the criterion. Simulation results show that the proposed approach is capable of separating independent sources from their post-nonlinear mixtures.

Separation of sources from one-dimensional mixture signals such as speech has been largely explored. However, two-dimensional sources (images) separation problem has only been examined to a limited extent. The reason is that ICA is a very general-purpose statistical technique, and it does not take the spatial information into account while separating mixture images. In this paper, we introduce Markov random field model to incorporate the spatial information into ICA. MRF is considered as a powerful tool to model the joint probability distribution of the image pixels in terms of local spatial interactions. An MRF-ICA based algorithm is proposed for image separation. It is successfully demonstrated on artificial and real images.

Through task decomposition and module combination, min-max modular support vector machines (M^3-SVMs) can be successfully used for difficult pattern classification task. M^3-SVMs divide the training data set of the original problem to several sub-sets, and combine them to a series of sub-problems which can be trained more effectively. In this paper, we explore the use of M^3-SVMs in multi-view face recognition. Using M^3-SVMs, we can decompose the whole complicated problem of multi-view face recognition into several simple sub-problems. The experimental results show that M^3-SVMs can be successfully used for multi-view face recognition and make the classification more accurate.

This paper presents a novel texture segmentation method using Bayesian estimation and neural networks. Multi-scale wavelet coefficients and the context information extracted from neighboring wavelet coefficients were used as input for the neural networks. The output was modeled as a posterior probability. The context information was obtained by HMT (Hidden Markov Trees) model. The proposed segmentation method shows performed better than ML (Maximum Likelihood) segmentation using the HMT model.

FastICA and CuBICA are two remarkable independent component analysis algorithms for dealing with blind signal separation problems. In this paper, we first present a novel ICA estimation algorithm, initialization constrained FastICA (IC-FastICA), through combining the technical merits of these two approaches. Then, a performance comparison study on these three approaches is conducted through the simulations on some standard benchmark data. The experimental results demonstrate that the IC-FastICA achieves higher performances on unmixing error and signal noise ratio while appreciably increasing computation cost.