There are many research activities in various recognition areas using high calibered computing power. Among many areas, the signature recognition and verification have more difficulties than any other recognition area because signature itself contains many problems caused by a variation of psychological status of signer and other environment. In the case of signature, therefore, it is important to extract the better parameters required for the higher verification ratio. In this paper, signature pressure is extracted and used as feature parameters to determine whether the input signature is ture or forgery, and then input signature is verified by fuzzy similarity method. As a result of appling the fuzzy similarity method to the recognition system it is proven that the system has by far better verification ratio about 10% than existing methods.

This paper proposes a image compression algorithm which stores and transmites image data efficiently. The proposed compression algorithm modify enhances compression rate by modified ZIG-ZAG Scanning in JPEG standard algorithm which is based on 2D-DCT. And the up-compatible method of proposed algorithm can solve compatible problem with JPEG that is cased by modified ZIG-ZAG Scanning. And this paper presentes a block diagram of hardware for real-time processing.

The major reason that reduces the performance of satellite communication channels is known as the nonlinearities of high power amplifiers(HPA's) which are usually modeled by memoryless nonlinear systems, in the channels. This paper proposes a nonlinear adaptive predistorter,which predistorts the input symbols such that the symbol passing through the nonlinear HPA preserves the desired constellation, and derives the region of convergence for the predistorter. The predistorter has capability of coping adaptively with any time-varying environment such as changes of operating points if HPA since the coefficients of the predistorter which has a form of truncated Tayler series are obtained by recursive optimization procedure utilizing QAM data sequence directely. Simulation results show that the proposed one with 16 QAM input has better performance than the conventional fixed predistorter with the same input.

The Lloyd-Max algorithm is an iterative scheme for design of the minimum mean square error quantizer. It is very simple in concept and easy to program into a computer. However its convergence and accuracy are primarily dependent upon the accuracy of the initial parameter. In this paper, a new initial parameter which converges to a specific value when the number of output levels becomes large is selected. And an estimator using curve fitting techique is suggested. In addition, performance of the proposed method is shown to be superior to that of the existing methods in accuracy and convergence.

A single channel adaptive noise canceling (ANC) technique with a recursive time delay estimator (RTDE) is presented for removing effects of additive noise on the speech signal. While the conventional method makes a reference signal for the adaptive filter using the pitch estimated on a frame basis from the input speech, the proposed method makes the reference signal using the delay estimated recursively on a sample-by-sample basis. As the RTDEs, the recursion formulae of autocorrelation function (ACF) and average magnitude difference function (AMDF) are derived. The normalized least mean square (NLMS) and recursive least square (RLS) algorithms are applied for adaptation of filter coefficients. Experimental results with noisy speech demonstrate that the proposed method improves the perceived speech quality as well as the signal-to-noise ratio and cepstral distance when compared with the conventional method.

In this paper, a low power and high speed flash Analog-to-Digital Converter using current-mode concept is proposed. Current-mode approach offers a number of advantages over conventional voltage-mode approach, such as lower power consumption small chip area improved accuracy etc. Rescently this concept was applied to algorithmic A/D Converter. But, its conversion speed is limited to medium speed. Consequently this converter is not applicable to the high speed signal processing system. This ADC is fabricated in 1.2um double metal CMOS standard process. This ADC's conversion time is measured to be 7MHz, and power consumption is 2.0mW, and differential nonlinearity is less than 1.14LSB and total harmonic distortion is -50dB. The active area of analog chip is about 350 x 550u$m^2$. The proposed ADC seems suitable for a single chip design of digital signal processing system required high conversion speed, high resolution small chip area and low power consumption.

An efficient approach for the computationtion of the optimum convergence factor is proposed for the LMS algorithm applied to a transversal FIR structure in this study. The approach automatically leads to an optimum step size algorithm at each weight in every iteration that results in a dramatic reduction in terms of convergence time. The algorithm is evaluated in system identification application where two alternative computer simulations are considered for time-invariant and time-varying system cases. The results show that the proposed algorithm needs not appropriate convergence factor and has better performance than AGC(Automatic Gain Control) algorithm and Karni algorithm, which require the convergence factors controlled arbitrarily in computer simulation for time-invariant system and time-varying systems. Also, itis shown that the proposed algorithm has the excellent adaptability campared with NLMS(Normalized LMS) algorithm and RLS (Recursive least Square) algorithm for time-varying circumstances.

A recursive on-line algorithm for orthogonal ARMA identification is proposed for linear MIMO systems with unknown parameters time delay and order. This algorithm is based on the Gram-Schmidt orthogonalization of basis functions, and extended to a recursiveform by using new functions of two dimensional autocorrelations and crosscorrelations of inputs and outputs. This proposed algorithm can also cope with slowly time-varying or order-varying systems. Various simulations reveal the performance of the algorithm.

In this paper, we present a new GPC algorithm for plants with known disturbances. Also, based on the above algorithm unidirectional and bidirectional predictive controllers have been proposed concerning plants with input couplings between subprocesses. These predictive control algorithms can be applied to many industrial processes as a multiloop controllers. Simulation results show that the proposed algorithm has improved performance and computing time compared to conventional GPC algorithm.

In this paper, an efficient algorithm is proposed which recognizes the mixed document consisting of the printed Korean/alphanumeric texts and graphic images. In the preprocessing step an input document is aligned if necessary by rotating it. We obtain the rotation angle using the Hough transform and align the input document horizontally. Then we separate graphic image parts from text parts by considering chain codes of connected components. We further separate each character using vertical and horizontal projections. In the recognition step Korean and alphanumeric characters are classified and each of them is recognized hierarchically using several features. In summary an efficient recognition algorithm for mixed documents is proposed and its performance is demonstrated via computer simulations.

The lapped orthogonal transform(LOT) has been recently proposed to alleviate the blocking effects in transform coding. The LOT is known to provide an improved coding gain than the conventional transform. In this paper, we propose a prefilter approach to design the LOT bases with the view of maximizing the transform coding gain. Since the nonlinear phase basis is inappropriate to the image coding only the linear phase basis is considered in this paper. Our approach is mainly based on decomposing the transform matrix into the orthogonal matrix and the prefilter matrix. And by assuming that the input is the 1st order Markov source we design the prefilter matrix and the orthogonal matirx maximizing the transform coding gain. The computer simulation results show that the proposed LOT provides about 0.6~0.8 dB PSNR gain over the DCT and about 0.2~0.3 dB PSNR gain over the conventional LOT [7]. Also, the subjective test reveals that the proposed LOT shows less blocking effect than the DCT.

In this work a scanning capacitance microscopy(SCaM) by stage driving is proposed and presented some of the experimental results.SCaM is a microscope which scans a surface of materials mechanically in two or two point five dimensions by a capacitance probe with a few tenth $\mu\textrm{m}$ ize tip, and display images of the surface shape or capacitive distribution. The present target of the SCaM is 0.1$\mu\textrm{m}$ resolution power which exceeds that of optical microscope. This will become a powerful tool for inspecting ULSI pattern etched by X-ray biological data etc. The experimental system is composed based on a VHD video disk which captures the capacitance changes of the video disk surface and converts it into video signal.

By using a concept of interrelation vector between line segments a new algorithm for partial shape recognition of two-dimensional objects is introduced. The interrelation vector which is invariant under translation rotation and scaling of a pair of line segments is used as a feature information for polygonal shape recognition. Several useful properties of the interrelation vector are also derived in relation to efficient partial shape recognition. The proposed algorithm requires only small space of storage and is shown to be computationally simple and efficient.

In this paper, we propose a method for facial feature extraction and recognition algorithm using neural networks. First we extract a face part from the background image based on the knowledge that it is located in the center of an input image and that the background is homogeneous. Then using vertical and horizontal projections. We extract features from the separated face image using knowledge base of human faces. In the recognition step we use the back propagation algorithm of the neural networks and in the learning step to reduce the computation time we vary learning and momentum rates. Our technique recognizes 6 women and 14 men correctly.

In this paper we investigate the hardware implementation of block matching algorithms (BMAs) for moving sequences. Using systolic arrays we propose a hardware architecture of a two-stage BMA using integral projections which reduces greatly computational complexity with its performance comparable to that of the full search (FS). Proposed hardware architecture is faster than hardware architecture of the FS by 2~15 times. For realization of the FS and two stage BMA modeling and simulation results using SPW and VHDL are also shown.

This paper presents a new method for reducing the blocking artifacts there by increasing the prediction gains of the block-based motion compensation keeping the amount of the motion information to be transmitted intact. The new method improves the motion compensation along the edges of moving objects by segmenting the motion field at the pixel resolution based on the model that the motion compensated image is the maximum a poseriori estimate of the current frame.

Recently the fuzzy control is widely used as a tool for constructing automatic control systems which can replace the manual operation of large-scale nonlinear plants. In most applications of the fuzzy control however it is hard to meet the requirement of the operation time. In some real-time control the fuzzy control scheme requires too much computing time for fuzzification inference and defuzzification. To reduce the computing time there may be two alternatives the development of a new operation algorithm and the design of high-speed fuzzy hardware. In this paper to solve the problem of reducing the fuzzy operation time we propose a new high-speed fuzzy hardware scheme which has merits of its generality and extensibility. Finally we verify the proposed fuzzy hardware.

In this paper, we develop a prototype of ITS(Intelligent Tutoring Systems) system: UNIX TUTOR. It is designed for the purpose of teaching the UNIX beginners the principal concepts of UNIX and the shell commands using the communication between the student and the system. UNIX TUTOR engages the student in a two-way conversation that is mixed-initiative dialogue and attempts to teach the student UNIX via the Socratic method of guided discovery and the Coaching method interchangeably. And the student model is based on both the overlay model and the buggy model together. Thus TUTOR aims at teaching the students effectively whose levels of learning are different using various explanations which are determined by the student model. Because the knowledge representation for UNIX-TUTOR is based on the frame structure and the production rules it is easy to represent the complicated constructs. UNIX TUTOR is implemented on the SPARC station using X/Motif and C for cp command among 10 ones which were selected.

The nearest neighbor classification rule is widely used because it is not only simple but the error rate is asymptotically less than twice Bayes theoretical minimum error. But the method basically use the whole training patterns as the reference vectors. so that both storage and classification time increase as the number of training patterns increases. LVQ(Learning Vector Quantization) resolved this problem by training the reference vectors instead of just storing the whole training patterns. But it is a heuristic algorithm which has no theoretic background there is no terminating condition and it requires a lot of iterations to get to meaningful result. This paper is to propose a new training method of the reference vectors. which minimize the given error function. The nearest neighbor network,the network version of the nearest neighbor classification rule is proposed. The network is funtionally identical to the nearest neighbor classification rule is proposed. The network is funtionally identical to the nearest neighbor classification rule and the reference vectors are represented by the weights between the nodes. The network is trained to minimize the error function with respect to the weights by the steepest descent method. The learning algorithm is derived and it is shown that the proposed method can adjust more reference vectors than LVQ in each iteration. Experiment showed that the proposed method requires less iterations and the error rate is smaller than that of LVQ2.

A two-phase neural network finds exact feasible solutions for a constrained optimization programming problem. The time-varying programming neural network is a modified steepest-gradient algorithm which solves time-varying optimization problems. In this paper, we propose a time-varying two-phase optimization neural network which incorporates the merits of the two-phase neural network and the time-varying neural network. The proposed algorithm is applied to system identification and function approximation using a multi-layer perceptron. Particularly training of a multi-layer perceptrion is regarded as a time-varying optimization problem. Our algorithm can also be applied to the case where the weights are constrained. Simulation results prove the proposed algorithm is efficient for solving various optimization problems.

In this paper, we propose the cross-section processing method which is simple in describing the SHGC of objects in a range image and which can describe the SHGC of occluded objects for the recognition of 3D objects. This method produces the cross-sections of an object along the assumed axis of the SHGC and describes the SHGC of the object by processing the produced cross-sections of the object using $\psi$ -S curves with invariant properties in position and size. Our method is simple in a process and can descirbe the SHGC of partially occluded objects because it uses range images with 3-D informations of objects without matching contours of objects with a model base. Thus it is a useful description method of a range image for the recognition of 3D objects shaped in SHGC form and we proved the usefulness of it in experiments.

Most of clustering methods usually employ a center or predefined shape of a cluster to assign the input data into the cluster. When there is no information about data set, it is impossible to predict how many clusters are to be or what shape clusters take. (the shape of clusters could not be easily represented by the center or predefined shape of clusters) Therefore, it is difficult to assign input data into a proper cluster using previous methods. In this paper, to overcome such a difficulty a cluster is to be represented as a collection of several subclusters representing boundary of the cluster. And membership functions are used to represent how much input data bllongs to subclusters. Then the position of the nearest subcluster is adaptively corrected for expansion of cluster, which the subcluster belongs to by use of a competitive learning neural network. To show the validity of the proposed method a numerical example is illustrated where FMMC(Fuzzy Min-Max Clustering) algorithm is compared with the proposed method.

Stochastic computation employs random pulse streams to represent numbers. In this paper, we study a new method to implement the number system which uses the ratio of the numbers of ones and zeros in the pulse streams. In this number system. if P is the probability that a pulse is one in a pulse stream then the number X represented by the pulse stream is defined as P/(1-P). We propose circuits to implement the basic operations such as addition multiplication and sigmoid function with this number system and examine the error characteristics of such operations in stochastic computation. We also propose a neuron model and derive a learning algorithm based on backpropagation for the 3-layered feedforward neural networks. We apply this learning algorithm to a digit recognition problem. To analyze the results, we discuss the errors due to the variance of the random pulse streams and the quantization noise of finite length register.