Catálogo de publicaciones - libros

In this paper, Umeyama’s eigen-decomposition approach to weighted graph matching problems is critically examined. We argue that Umeyama’s approach only guarantees to work well for graphs that satisfy three critical conditions: (1) The pair of weighted graphs to be matched must be nearly isomorphic; (2) The eigenvalues of the adjacency matrix of each graph have to be single and isolated enough to each other; (3) The rows of the matrix of the corresponding absolute eigenvetors cannot be very similar to each other. For the purpose of matching general weighted graph pairs without such imposed constraints, we shall propose an approximate formula with a theoretical guarantee of accuracy, from which Umeyama’s formula can be deduced as a special case. Based on this approximate formula, a new algorithm for matching weighted graphs is developed. The experimental results demonstrate great improvements to the accuracy of weighted graph matching.

This paper presents a robust real-time face alignment algorithm based on Active Appearance Models(AAMs). Fitting an AAM to an image is considered to be a problem of minimizing the error between the input image and the closest model instance. If the input image is far from the model space, the fitting process will fail. This can always occur in application because of illumination variation. So, building a good appearance space is very important. We propose a weighted cost function which can incorporate intensity and edgeness of an image into AAMs framework. To achieve high performance, Active Appearance Models proposed by Iain Matthews is employed.

This paper presents an speech information hiding model for transmitting secret speech through subliminal channel covertly for secure communication over PSTN (Public Switched Telephone Network) or VoIP (Voice over Internet Protocol). This model’s main statement is that the embedding operation of a secure communication system should work indeterminate from the attacker’s point of view. This model for secure communication based on the technique of information hiding has more severe requirements on the performances of data embedding than watermarking and fingerprinting in the aspects of real time, hiding capacity, and speech quality. Experiments show that this model meets the requirement of secure communication and suits for practical application of covert communication. The security analysis of this model by means of information theory and actual test proved that it is theoretically and practically secure. This information theory based model can be commonly used to help design a system of speech secure communication with different coding schemes.

This paper presents an approach for speech information hiding based on G.721 scheme. This approach proposes an Analysis-By-Synthesis (ABS)-based speech information hiding and extracting algorithm, is called ABS algorithm, which form the theoretical basis for designing a secure speech communication system. The ABS algorithm adopts a speech synthesizer in a speech coder. Speech embedding and coding are synchronous, i.e. fusing of secret speech information data into speech coding. Dynamic secret speech information data bits can be embedded into original carrier speech data, with high efficiency in steganography and good quality in output speech. This method is superior to available classical algorithms on hiding capacity and robustness. This paper implements the proposed approach based on speech coding scheme G.721 and the experiments show that this approach meets the requirements of information hiding, satisfies the constraints of speech quality for secure communication, and achieves high hiding capacity of 1.6Kbps with an excellent speech quality and complicating speakers’ recognition.

Based on the research of MDS array code of size n(n in distributed storage system, in this paper, we present a novel encoding scheme called the and prove that the column distance of the F-code is 3, i.e. F-code is a MDS array code given that odd number n is greater than 3 and does not include factor 3. And we also implement a novel decoding algorithm of the F-code. The algorithm only needs two decoding chains in each linear equation group and is able to recover all unknown symbols on two erasure columns. The analysis of F-code shows that our method extends the range of number n in n×n MDS array code and gets lower/reduction algorithmic complexity. Therefore, the reliability of a distributed storage system that features the F-code can be effectively reinforced.

A new method of motion deadlock resolving by using fuzzy decision in robot soccer games is proposed in this paper. For the reasons of complex competition tasks and limited intelligence, soccer robots fall into motion deadlocks in many conditions, which is very difficult for robots to decide whether it is needed to retreat for finding new opportunities. Based on the analysis of human decision for dealing with these kinds of motion deadlocks, the fuzzy decision method is introduced in this paper. Then, fuzzy rules based deadlock resolving system is designed according to relative positions and orientations among robots and the ball in local regions. Lots of experiments by human experts and the fuzzy controller are implemented for comparison. Experimental results show that the method proposed is reasonable and efficient for motion deadlock resolving in most conditions for real soccer robot games.

In this paper, we propose the novel feature extraction method, called the Polar wavelet, which can improve the search performance for locally distributed time series data. Among various feature extraction methods, the Harr wavelet has been popularly used to extract features from time series data. However, the Harr wavelet does not show the good performance for sequences of similar averages. The proposed method uses polar coordinates which are not affected by averages and can reduce the search space efficiently without false dismissals. The experiments are performed on real temperature dataset to verify the performance of the proposed method.

In this paper medical applications of supervised and unsupervised neural networks image processing algorithms are presented and discussed by means of quantitative experimental results in the field of radiotherapy. The investigated case study concerns the problems and the consequent solutions referred to the two phases of the treatment plan necessary after the quadrantectomy of a cohort of patients affected by breast cancer.

Recognition of road guidance signs is an important issue for developing driving assistance systems. One of major problems in this field is recognition of junction structure from road signs. In the proposed paper both detection and recognition problems are presented. Detection method based on color and shape properties of sign allows detecting signs in various lighting and weather conditions. Using interlaced images for most time-consuming operations and full-resolution images for final result reduces computation time without loss of quality. Recognition method is based on decomposition of guidance signs into principal components and representation of arrow region as a graph. Path extraction uses finite automate methodology which in order to recognize all possible paths to pass the junction. Proposed method showed processing speed about 10 fps and can be used in real-time applications.