Shao-Han Liu

Vector quantization in DCT domain using fuzzy possibilistic c-means based on penalized and compensated constraints

In this paper, fuzzy possibilistic c-means (FPCM) approach based on penalized and compensated constraints are proposed to vector quantization (VQ) in discrete cosine transform (DCT) for image compression. These approaches are named penalized fuzzy possibilistic c-means (PFPCM) and compensated fuzzy possibilistic c-means (CFPCM). The main purpose is to modify the FPCM strategy with penalized or compensated constraints so that the cluster centroids can be updated with penalized or compensated terms iteratively in order to 4nd near-global solution in optimal problem. The information transformed by DCT was separated into DC and AC coe5cients. Then, the AC coe5cients are trained by using the proposed methods to generate better codebook based on VQ. The compression performances using the proposed approaches are compared with FPCM and conventional VQ method. From the experimental results, the promising performances can be obtained using the proposed approaches. ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.

A competitive continuous Hopfield neural network for vector quantization in image compression

Vector quantization is a system in which a distortion function is minimized for multidimensional optimization problems. The purpose of such a system is data compression. In this paper, a parallel approach using the competitive continuous Hopfield neural network (CCHNN) is proposed for the vector quantization in image compression. In CCHNN, the codebook design is conceptually considered as a clustering problem. Here, it is a kind of continuous Hopfield network model imposed by the winner-take-all mechanism, working toward minimizing an objective function that is defined as the average distortion measure between any two training vectors within the same class (within-class). It also forward maximizes an objective function defined as the average distortion measure between any two training vectors in separate classes (between-class). For an image of n training vectors and c objects of interest, the proposed CCHNN would consist of n c neurons. Each neuron (or training vector) occupies l l components of a training vector. In the experimental results, the proposed method shows more promising results after convergence than the generalized Lloyd algorithm. # 1999 Elsevier Science Ltd. All rights reserved.

Classification of multispectral images through a rough-fuzzy neural network

A new fuzzy Hopfield-model net based on rough-set reason- ing is proposed for the classification of multispectral images. The main purpose is to embed a rough-set learning scheme into the fuzzy Hopfield network to construct a classification system called a rough-fuzzy Hopfield net (RFHN). The classification system is a paradigm for the implementation of fuzzy logic and rough systems in neural network ar- chitecture. Instead of all the information in the image being fed into the neural network, the upper- and lower-bound gray levels, captured from a training vector in a multispectal image, are fed into a rough-fuzzy neuron in the RFHN. Therefore, only 2/N pixels are selected as the training samples if an N-dimensional multispectral image was used. In the simu- lation results, the proposed network not only reduces the consuming time but also reserves the classification performance.

Classification of multispectral images based on a fuzzy-possibilistic neural network

In this paper, a new Hopfield-model net based on fuzzy possibilistic reasoning is proposed for the classification of multispectral images. The main purpose is to modify the Hopfield network embedded with fuzzy possibilistic C-means (FPCM) method to construct a classification system named fuzzy-possibilistic Hopfield net (FPHN). The classification system is a paradigm for the implementation of fuzzy logic systems in neural network architecture. Instead of one state in a neuron for the conventional Hopfield nets, each neuron occupies 2 states called membership state and typicality state in the proposed FPHN. The proposed network not only solves the noise sensitivity fault of Fuzzy C-means (FCM) but also overcomes the simultaneous clustering problem of possibilistic C-means (PCM) strategy. In addition to the same characteristics as the FPCM algorithm, the simple features of this network are clear potential in optimal problem. The experimental results show that the proposed FPHN can obtain better solutions in the classification of multispectral images.

A shortest-path network problem using an annealed ant system algorithm

This paper investigates a shortest-path network problem using an annealed ant system algorithm, in which an annealing strategy is embedded to calculate the probabilities to decide which path the ants will select next. The shortest-path problem determines the shortest route between a source and a destination in a transportation-network topology. In this approach, according to the concrete problems of shortest routing, we construct two globally optimizing annealed ant algorithms that are the concentrated model and distributed model. The concentrated model (CM) means all ants are initially concentrated in the source node while all ants randomly select a node except the destination as their starting point initially and at least one must appear in the source node for the distributed model (DM). The experimental results show that the proposed annealed ant algorithm with roulette wheel selection can obtain better performance than that generated by the traditional ant strategy with/without roulette wheel selection.

Determining 2D notch SIFs by the image-correlation method

This study evaluates the two-dimensional (2D) stress intensity factors (SIFs) of a sharp V-notch using the image-correlation experiment and least-square method for isotropic materials. First, the Williams eigenfunction and complex displacement function approach are deduced into a least-square form, and then displacement fields from the image-correlation experiment are substituted into the least-square equation to evaluate the 2D SIFs. Compared with the SIFs from finite element and body force methods, the least-squares method can be used to calculate SIFs more accurately, if more than two displacement terms are included. The SIFs calculated from this least-squares method are not sensitive to the maximum or minimum radius of the area from which data is included. The major advantage of the proposed method is that the procedure is simple and systematic, so it can be applied to any finite element or experimental methods that obtain displacement fields.

A COMPENSATED FUZZY HOPFIELD NEURAL NETWORK FOR CODEBOOK DESIGN IN VECTOR QUANTIZATION

In this paper, a new Hopfield-model net called Compensated Fuzzy Hopfield Neural Network (CFHNN) is proposed for vector quantization in image compression. In CFHNN, the compensated fuzzy c-means algorithm, modified from penalized fuzzy c-means, is embedded into Hopfield neural network so that the parallel implementation for codebook design is feasible. The vector quantization can be cast as an optimal problem that may also be regarded as a minimization of a criterion defined as a function of the average distortion between training vector and codevector. The CFHNN is trained to classify the divided vectors on a real image into feasible class to generate an available codebook when the defined energy function converges to near global minimum. The training vectors on a divided image are mapped to a two-dimensional Hopfield neural network. Also the compensated fuzzy c-means technique is used to update the quantization performance and to eliminate searching for the weighting factors. In the context of vector quantization, each training vector on the divided image is represented by a neuron which is fully connected by the other neurons. After a number of iterations, neuron states are refined to reach near optimal result when the defined energy function is converged.

Character recognition with CMAC on field programmable gate array

We proposed a cerebellar model arithmetic computer (CMAC) neural network to characters recognition on an FPGA architecture. The CMAC has many advantages in terms of speed of operation based on LMS training. Its ability realizes arbitrary nonlinear mapping and a fast practical hardware implementation. This work presents CMAC hardware that is about 35 times faster than that by the software executed on the conventional processor. In the experimental results, the CMAC is shown that it can clearly distinguish 94 characters with a size of 8/spl times/8 pixels though there are some noise pixels in a character.

A Boundary Representation Based on Polygonal Approximation with an Ant Colony Algorithm for the Two-Dimensional Objects

This paper investigates polygonal approximation in boundary representation for two-dimensional objects using an ant colony algorithm. Ant colony algorithm is a newly optimization algorithm for the field of stochastic researching recently. In the proposed approach, in according with the pheromone strength and the arc-to-chord distance for a curve, we construct an optimizing Ant Colony Algorithm (ACA) that based on the ability of ants to find the optimal dominant points in a curve between the source and destination. The experimental results show that the proposed ACA with the roulette wheel selection can obtain better performance than that generated by the conventional polygonal approximation methods.

Adhesion status of platelets under flow conditions by digital image processing

Abstract The kinetics and distribution of platelet deposition on protein‐coated surfaces were studied in an in vitro flow system. When fluorescence‐labeled platelets in whole blood flow through a flow chamber that is composed of a glass plate, platelets will deposit onto the protein‐coated surface in a time‐and location‐dependent manner. A combination of fluorescence video microscopy and a digital image processing system allowed us to systematically study thrombosis kinetics under various flow conditions, with different biomaterials and forms of blood. In this paper, the image processing and pattern recognition techniques have been developed to quantify adhered platelets from numerous video frames. The dynamic adhesion status (newly attached, staying, detached) has been calculated automatically by software developed on an IBM/PC/386. These morphometric parameters as a function of flow rate and type of biomaterials can be determined experimentally.