In this paper, we propose an efficient two-step pattern matching method which promises shape distortion-tolerant recognition of handwritten of handwritten Hangul syllables. In the first step, nonlinear shape normalization is carried out to compensate for global shape distortions in handwritten characters, then a preliminary classification based on simple pattern matching is performed. In the next step, nonlinear pattern matching which achieves best matching between input and reference pattern is carried out to compensate for local shape distortions, then detailed classification which determines the final result of classification is performed. As the performance of recognition systems based on pattern matching methods is greatly effected by the quality of reference patterns. we construct reference patterns by combining the proposed nonlinear pattern matching method with a well-known averaging techniques. Experimental results reveal that recognition performance is greatly improved by the proposed two-step pattern matching method and the reference pattern construction scheme.

The incremental parsing techniques are essential part of language-based environment which allow incremental construction of programs, as reparsing of modified part of the program alone enhances the system performance. A number of methods and algorithms have been proposed for incremental parsing and for the construction of incremental parsers. The purpose of this paper is to review the earlier incremental parsing algorithms which are too expensive in both time and space, and to present an efficient incremental LR parsing algorithm which is more efficient than the previous ones. In this paper, we generate automatically an extended LR parsing tables which allow the nonterminal symbols as well as terminal symbols for the input, and apply them to our efficient incremental LR parsing algorithm. Using these extended LR parsing tables. we suggest several methods to reduce its memory spaces and parsing steps as well. The algorithms described here were implemented in C language and YACC on a SUN UNIX OS, and were tested with several grammars and strings.

For the efficient recognition of Korean Alphabets, we proposed the tree structure algorithm which was based on K-tuple NRF-SDF using representative images as training images. Representative images consisted of ECP-SDF images of several consonants or vowels. To reduce the effect of sidelobe in the output correlation plane, we used the representative images as training images and obtained the elements of a vector inner product matrix using the peak value of AMPOF correlation of training images with one another. The proposed algorithm consisted of three main-step containing several substeps. In filter synthesis of each step, representative images were used as training images in the first and the second main-step and each consonant or vowel was used as training images in the third main-step. The performance of this algorithm is demonstrated by computer simulation and optical experiment.

A 1-chip high quality binarizing VLSI image processor (which has 8 bit ADC. 6 bit flash ADC, 15K standard cell, and 1K word ROM) based on 10 MIPS 16 bit DSP is implemented for FAX. This image processor(IP) performs image pre-processing. image quality improvement in copying and sending mode, and mixed image processing based on the fuzzy theory. And smoothing in sub-scan direction is applied for normal receiving mode data so the received data is enhanced like fine mode data. Each algorithm is processed with the same type of image processing window and 2-D image processing is implemented with a 1-D line buffer. The fabricated chip is applied to a FAX machine and image quality improvement is verified.

In this paper, Micro-Genetic algorithms($\mu$-GAs) is proposed on a minimum-time path planning for robot manipulator. which is a kind of optimization algorithm. The minimum-time path planning, which can allow the robot system to perform the demanded tasks with a minimum execution time, may be of consequence to improve the productivity. But most of the methods proposed till now suffers from a significant computation burden and can`t often find the optimaul values. One way to overcome such difficulties is to apply the Micro-Genetic Algorithms, which can allow to find the optimul values, to the minimum-time problem. This paper propose an approach for solving the minimum-time path planning by using Micro-Genetic Algorithms. The effectiveness of the proposed method is demonstrated using the 2 d.o.f plannar Robot manipulator.

A recursive on-line algorithm with orthogonal ARMA identification is proposed for linear systems with unkonwn time delay, order, and parameters. The algorithm is based on the Gram-Schmidt orthogonalization of basis functions, and extendedto recursive form by using two dimensional autocorrelations and crosscorrelations of input and output with constant data length. The proposed algorith can cope with slowly time-varying or order-varying delayed system. Various simulations reveal the performance of the algorithm.

The multiple valued logic(MVL) circuit with ROM type using current mode CMOS is presented in this paper. This circuit is composed of the multiple valued-to-binary(MV/B) decoder and the selection circuit. The MV/B decoder decodes the single input multiple valued signal to N binary signal, and the selection circuits is composed N$\times$N array of the selecion cells with ROM types. The selection cell is realized with the current mirror circuits and the inhibit circuits. The presented circuit is suitable for designing the circuit of MVL functions with independent variables, and reduces the number of selection cells for designing the circuit of symmetric MVL functions as many as {($N^2$-N)/2}+N. This circuit possess features of simplicity. expansibility for array and regularity, modularity for the wire routing. Also, it is suitable for VLSI implementation.

In this paper, a segmentation-based coding technique as applied to video sequences is proposed. A proposed method separates an image into contour and texture parts, then the visually-sensitive contour part is represented by chain codes and the visually-insensitive texture part is reconstructed by a representative motion vector of a region and mean of the segmented frame difference. It uses a change detector to find moving areas and adopts variable blocks to represent different motions correctly. For better quality of reconstructed images, the displaced frame difference between the original image and the motion compensated image reconstructed by the representative motion vector is segmented. Computer simulation with several video sequences shows that the proposed method gives better performance than the conventional ones in terms of the peak signal to noise ratio(PSNR) and compression ration.

This paper presents recognition of raised alphanumeric markings on rubber tires for their automatic classification. Raised alphanumeric markings on rubber tires have different characteristics as compared to those of printed characters. In the preprocessing step, we first determine the rotation angle using the Hough transform and align markings, then separate each character using vertical and horizontal projections. In the recognition step, we use several features such as width of a character, cross point, partial projection, and distance feature to recognize characters hierarchically. The computer simulation result shows that the proposed system can be successfully applied to the industrial automation of rubber tires classification.

In this paper, we propose a subband pyramid image coding scheme that uses ECVQ (ntropy Constrained Vector Quantizer). In subband pyramid image coding, each subband can be encoded with a coder matched to the statistics of that particular subband, and available versions of the original image at different resolution are easily obtained. ECVQ, aiming at the minimization of the distortion for a fixed entropy of the quantizer output, is well combined with the subband pyramid image coding which yields high compression ratio and good image quality. The optimum bit allocation to each subbands corresponds to the points where the individual distortion rate curves are of particular slope, weighted to the number of samples in that subband, in designing ECVQ.

Transform image coding using DCT is proved to be efficient in the absence of channel error but its performance degrades rapidly over noisy channel. In this paper, in the case of appling bit selcetive error correction coding that protects some significant bits in a codeword, an efficient allocation method of imformation bits and additive redundancy bits used for quantization and error correction coding respectively under constant transmission bit rate is proposed, and its performance is analyzed. As a result, without increasing trasmission bit rate, PSNR can be improved up to 7~8 [dB] below bit error rate $10^2$ and the image without blocking effect caused by bit error resulted from channel noise can be recostructed.

A new merging technique is adopted for combining rectangular parallelepipes produced by 2-D rectangular code into more intuitive 30D volume elements. Rectangular parallelepiped codes (RP codes) can be used in volume-based representation of a three-dimensional object. We proposed more regularity-conserving 2-D rectangular coding scheme to merge rectangular cells represented by RP codes in three-dimensional space. After being constructed from modified 2-D rectangular code, 3-D RP codes are merged in the two orthogonal directions using new merging algorithm. The shape of merged 3-D object reconstructed by proposed algorithm is shown to be much closer to the original object shape than that of conventional RP codes. The storage requirement of merged object can be also reduced.

Roberige had proposed a method of linear approximation using strips. This method is known to be fast relatively, and the distance between the given curve and calculated straight line is not greater than an allowable error. But this method generates many break points. Later, Leng and Yang proposed the dynamic strip algorithm with the rotation of strips which reduced the number of greak points at the sacrifice of the speed. The method using the seperate rotation of two strip lines proposed in this paper is faster than Leng and Yang's but generates almost the same number of break points as Leng and Yang's.

Hardware realization of a digital 2-D homomorphic filter for real time contrast enhancement of video signal is presented. In homomorphic filter, logarithmic and exponential conversion used the memory lookup table method and because the hardware is implemented by multiplierless TTL devices, it can be designed to specific IC. The contrast gain can be controlled externally and the transfer function of homomorphic filter can be easily varied by the change of lookup table memory data.

In this paper, a new texture feature extraction method for texture image classification is proposed. The proposed method is a modified texture spectrum method. It uses local weighting averaged texture unit, that is, the neighbor pixels are weithted and averaged in 4-direction and the calculated values are compared with center pixel to find texture units. The proposed method has only 81 texture units and these units are really good features for texture classification. The proposed method is applied to vegetable images and Blodatz album images and compared with several conventional methods for the feature extraction time and the recognition rate.

An inference network is proposed as a tool for bidirectional approximate reasoning. The inference network can be designed directly from the given fuzzy data(knowledge). If a fuzzy input is given for the inference netwok, then the network renders a reasonable fuzzy output after performing approximate reasoning based on an equality measure. Conversely, due to the bidirectional structure, the network can yield its corresponding reasonable fuzzy input for a given fuzzy output. This property makes it possible to perform forward and backward reasoning in the knowledge base system.

This paper proposes an efficient method for improving the training performance of the neural network by using a hybrid of a stochastic approximation and a backpropagation algorithm. The proposed method improves the performance of the training by appliying a global optimization method which is a hybrid of a stochastic approximation and a backpropagation algorithm. The approximate initial point for a stochastic approximation and a backpropagation algorihtm. The approximate initial point for fast global optimization is estimated first by applying the stochastic approximation, and then the backpropagation algorithm, which is the fast gradient descent method, is applied for a high speed global optimization. And further speed-up of training is made possible by adjusting the training parameters of each of the output and the hidden layer adaptively to the standard deviation of the neuron output of each layer. The proposed method has been applied to the parity checking and the pattern classification, and the simulation results show that the performance of the proposed method is superior to that of the backpropagation, the Baba's MROM, and the Sun's method with randomized initial point settings. The results of adaptive adjusting of the training parameters show that the proposed method further improves the convergence speed about 20% in training.

An analog neural network circuit of rhigh density integration is introduced. It's prototype chip is designed in 3 by 3 mm2 die. It uses only one MOSFET to implement a synapse. The number of synapses per neuron can be expanded by cascading several chips. The influence of nonlinearity in synapses is analyzed. A formalization of the back propagation which can be applied to this circuit is shown. Some simulation results are shown and disscussed.

This paper presents the algorithm of an on-line cursive Hangeul recognition by means of the DP matching with structural information. In the proposed method, the penalty represents the similarity between input character and reference character. The penalty is dynamically computed with types of strokes, directions of strokes, and the length rations of fragment during the process that searchs for an optimal path using 3$\times$3 matrix. As the result, this method can be exactly matched for even greatly deformed characters. The computer simulation shows that the proposed algorithm can be utilized in recognizing cursive Hangeul as well as correct Hangeul.

In this study, we present unvoiced consonant recognition system using characteristic parameters of the phoneme of the each syllable. For the recognition, the characteristic parameters on the time domain such as ZCR, total energy of the consonant region and half region energy of the consonant region, and those on the frequency domain such as the frequency spectrum of the transition region are used. The objective unvoiced consonants in this study are /ㄱ/,/ㄷ/,/ㅂ/,/ㅈ/,/ㅋ/,/ㅌ/,/ㅍ/ and /ㅊ/. Each characteristic parameter of two regions extracted from these segmented unvoiced consonants are used for each recognition system of the region, independently, And complementing two outputs of each other system, the final output is to be produced. The recognition system is implemented using MLP which has learning ability. The recognition simulation results for 112 unvoiced consonant samples are that average recognition rates are 96.4$\%$ under 80$\%$ learning rates and 93.7$\%$ under 60$\%$ learning rates.