In this paper, we observe that the design of optimal weighted order statistic(WOS) filters minimizing the mean absolute error criterion can be though of as a two-class linear classification problem. Based on this observation, the perceptron algorithm is applied to design WOS filters. It is shown, through experiments, that the perceptron algorithm can find optimal or near optimal WOS filters in practical situations.

This paper presents a new concept of a reconfigurable manipulator system which adjusts its mechanical structure to suit the kinematic characteristics of a given task. A highly redundant manipulator designed as a general purpose manipulator needs to be reconfigured for a specific task. A general task can be decomposed of motion and force components with different control requirements: either gross motion control or fine motion control. Each of these task components are distributed to each part of the manipulator based on the control requirements and the structure of the manipulator. Through the reconfiguration, a redundant manipulator is decomposed into two local arms, and the kinematic characteristics of each local arm is adjusted to suit the assigned task. The reconfigured redundant manipulator has two local arms well-configured for the local tasks and cooperating in serial for a given task. This globally enhances the performance of a redundant manipulator to execute a specific task. The simulation results are shown.

The white-balance system for color tevision is characterized by 5 input-5 output nonlinear process. A design strategy of fuzzy control rules is treated in which it can be adopted to the white balance adjustment for color television. A fuzzy rule based on an expert's knowledge is constructed, and then a multivariable fuzzy control rule is designed. Since human has just two hands, he can manipulate two variables simutaneously. In case when the process to be controlled has more than three control variables, expert's control rule is much different from the multivariable control rule. A multivariable fuzzy control rule is constructed by utilizing the expert' knowledge and rough relations between input and output variables, and its usefulness is shown by experiments.

An algorithm for MMADF(modified multiplication-free adaptive filter) which need not to multiplication arithmatic operation is proposed to improve the performance of FSE (fractionally spaced equalizer) which reduce the ISI(intersymbol interference) in signal transfer channel. The input signals are quantized using DPCM and the reference signals is processed using a first-order linear prediction filter. The convergence properties of Sign. MADF and M-MADF algorithm for updating of the coefficients of a FIR digital filter of the fractionally spaced equalizer (FSE) are investigated and compared with one another. The convergence properties are characterized by the steady state error and the convergence speed. It is shown that the convergence speed of M-MADF is almost same as Sign algorithm and is faster than MADF in the condition of same steady state error. Especially it is very useful for high correlated signals.

In this paper, a hierarchical motion vector estimation algorithm using pyramidal structure is proposed. Using a smaller measurement window at each level of a pyramid than that of the conventional scheme, the proposed algorithm, based on the TSS(three step search), reduces the computational complexity greatly with its performance comparable to that of the TSS. By increasing the number of cnadidate motion vectors which are to be used as the initial search points for motion vector estimation at the next level, the performance improves further. Then the computational complexity of the proposed hierarchical algorithm depends on the number of candidate motion vectors, with its PSNR (peak signal to noise ratio) ranging between those of the TSS and the full search method. The simulation results with two different block sizes and various test sequences are given and its hardware implementation is also sketched.

Extended reduced difference pyramid (ERDP) is proposed for lossless progressive image transmission, which is based on a new transform called rounded-transform(RT). The RT is a nonlinear and reversible transform of integers into integers utilizing two kinds of the rounding operations such as round up and down. The ERDP can be obtained from an N-poing RT or a series of RTs of both. For the performance evaluation, the entropy of the difference images to be transmitted is used as a lower bound transmission rate. Two examples of the ERDP can be easily shown, which is more effective in the entropy than the ordinary RDP.

Stereo vision is useful in obtaining three dimensional depth information from two images taken from different view points. Neural network modeling for stereo matching, the key step in stereo vision, is defined by an energy function satisfying with three constraints proposed by Marr and Poggio. Stereo matching is then carried out through the network to find minimum energy corresponding to the optimized solution of the problem. An algorithm for stereo matching using an analog neural network is presented here. The network can reduce errors in initial state an early iteration steps by adoption of continuous sigmoid function in stead of binary state. The experimental results show good matching performance for sparse random dot stereogram and real image.

ECG pattern was classified using a back-propagation neural network. An improved feature extractor of ECG is proposed for better classification capability. It is consisted of preprocessing ECG signal by an FIR filter faster than conventional one by a factor of 5. QRS complex recognition by moving-window integration, and peak extraction by quadratic approximation. Since the FIR filter had a periodic frequency spectrum, only one-fifth of usual processing time was required. Also, segmentation of ECG signal followed by quadratic approximation of each segment enabled accurate detection of both P and T waves. When improtant features were extracted and fed into back-propagation neural network for pattern classification, the required number of nodes in hidden and input layers was reduced compared to using raw data as an input, also reducing the necessary time for study. Accurate pattern classification was possible by an appropriate feature selection.

In this paper, we propose a stroke matching method for the off-line recognition of handprinted Hangul. In this method, the preprocessing steps such as position normalization, contour tracing and thinning are carried out first. Then, after extracting features such as the firection component distribution of contour, the direction component distribution of skeleton, and the distribution of structural feature points, strokes are extracted and matched based on the midpont distribution of the direction and the length of each stroke. In order to reduce the recognition time, a preliminary classification based on the direction component distribution features of the contour is performed. In order to domonstrate the performance of the proposed method, experiments with 520 most frequently used Hangul were performed, and 90.7% of correct recognition rate and 0.46second of recognition time per one character has been obtained. This results reveal that the proposed method can absorb effectively the noise in input character and the variations of stroke slant.

In this paper, we present a VLSI chip design of Multi-layer Bidirectionsl Associative Memory with good error-correction performance. The MBAM neural chip utilizes inner product implementation schems with binary storage and analog calculation.. Multi-layer can be constructed by direct cascading of these chips, and the number of neurons is expandable by parallel connection of these chips. We made proto-type chips and interface board to test the expansion. Currently the Chip has 32 input nodes, 32 output nodes, and can store up to 48 patterns, 32x48x2 SRAMs are included in the chip.

In previous paper we proposed a multi-layer bidirectional associative memory (MBAM) which is an extended model of the bidirectional associative memory (BAM) into a multilayer architecture. And we showed that the MBAM has the possibility to have binary storage for easy implementation. In this paper we present a MOdified MBAM(MOMBAM) with high performance compared to MBAM and multi-layer perceptron. The contents will include the architecture, the learning method, the computer simulation results for MOMBAM with MBAM and multi-layer perceptron, and the convergence properties shown by computer simulation examples.. And we will show that the proposed model can be used as classifier with a little restriction.

In this paper, noticing the point of human's ability which appropriately cope with vague conditions, we design fuzzy traffic signal light controller similar to human's distinction ability and decide the optimum cycle most suited to any traffic junction using fuzzy identification algorithm. In this study, for the control output decision process we design fuzzy controller better than electronic vehicle actuated controller in performance. We propose the cycle decision method which is not limited by the variance of traffic junction vehicle number through overcoming the limit of Webster's method which is adopted by the fixed cycle controller. Simulated experimental results show that fuzzy controller and fuzzy identification algorithm are better than the existing electronic vehicle actuated controller and fixed cycle controller in delay time per vehicle.