In previous papers the building of hierarchical networks made up of components using fuzzy rules was presented. It was demonstrated that this approach could be used to construct networks to solve classification problems, and that in many cases these networks were computationally less expensive and performed at least as well as existing approaches based on feedforward neural networks. It has also been demonstrated how this approach could be extended to real-valued problems, such as function approximation and time series prediction. This paper investigates the problem of choosing the best network for real-valued approximation problems. Firstly, the nature of the network parameters, how they are interrelated, and how they affect the performance of the system are clarified. Then we address the problem of choosing the best values of these parameters. We present two model selection tools in this regard, the first using a simple statistical model of the network, and the second using structural information about the network components. The resulting network selection methods are demonstrated and their performance tested on several benchmark and applied problems. The conclusions look at future research issues for further improving the performance of the clustering network.

The present paper is concerned with an algorithm for the minimization of multiple-valued input, binary-valued output functions. The algorithm is an extension to muitiple-valued logic of an algorithm for the minimization of ordinary single-output Boolean functions. It is based on a local covering approach. Basically, it uses a "divide and conquer" technique, consisting of two steps called expansion and selection. The present algorithm preserves two important features of the original one. First, a lower bound on the number of prime implicants in the minimum cover of the given function is furnished as a by-product of the minimization. Second, all the essential primes of the function are identified and selected during the expansion process. That usually improves efficiency when handling functions with many essential primes. Results of a comparison of the proposed algorithm with the program ESPRESSO-IIC developed at Berkeley are presented.

We present a recursive algorithm for constructing linear discrete-time systems which interpolate the desired 1st-and 2nd-order information. The recursive algorithm constructs a new system and connects it to the previous system in the cascade form every time new information is added. These procedures yield a practical realization of all the interpolants.

In this paper, we propose a new method of dialogue data collection which can be used to evaluate modules of a spoken dialogue system. To evaluate the module, it is necessary to use suitable data. Human-human dialogue data have not been appropriate to module evaluation, because spontaneous data usually include too much specific phenomena such as fillers, restarts, pauses, and hesitations. Human-machine dialogue data have not been appropriate to module evaluation, because the dialogue was unnatural and the available vocabularies were limited. Here, we propose 'Hybrid method' for the collection of spoken dialogue data. The merit is that, the collected data can be used as test data for the evaluation of a spoken dialogue system without any modification. In our method a human takes the role of some modules of the system and the system, also, works as the other part of the system together. For example, humans works as the speech recognition module and the dialogue management and a machine does the other part, response generation module. The collected data are good for the evaluation of the speech recognition and the dialogue management modules. The reasons are as follows. (1) Lexicon: The lexicon was composed of limited words and dependent on the task. (2) Grammar: The intention expressed by the subjects were concise and clear. (3) Topics: There were few utterances outside the task domain. The collected data can be used test data for the evaluation of a spoken dialogue system without any modification.

The properties of the optical-microwave mixing process are investigated in detail. To describe these processes a new approach, the parametric method is introduced which provides a better description of the mixing phenomenon. The paper presents new experimental results on and new theoretical analysis methods for the optical-microwave mixing process and also for its dynamic behavior. The dynamic properties are very important in many applications when the light is intensity modulated by a high frequency or high bit rate signal. A remarkable decrease is observed in the mixing product with increasing optical modulation frequency. There are two reasons for it: the time constant exhibited by the depletion region between the substrate and the epitaxial layer and the optically induced substrate current which is increasing with the modulation frequency and doesn't contribute to the mixing effect. Understanding the optical-microwave mixing process provides new solutions for many applications. For example the optical-microwave mixing techniques offers several advantages in case of optical reception. In the detection process the modulation signal can be transposed to an intermediate frequency band (instead of the baseband) making possible a lower noise reception in a wider band. Another important and advantageous application is in the reception of subcarrier modulated optical signals.

In this communication we propose to solve the problem of reconstruction from limited data using a statistical regularization method based on a Bayesian information criterion. The minimization of the Bayesian information criterion, which is used here as an objective index to measure the goodness of an estimate, gives the optimum value of the smoothing parameter. By doing so, we could reduce the inversion problem to a simple minimization of a one-variable nonlinear function. The application of such a technique overcomes the nonuniqueness of the solution of the ill-posed problem and all shortcomings of many iterative methods. In the light of simulation and application to real data, we propose a slight modification to the Bayesian information criterion to reconstruct the wave energy distribution at the ionospheric base from the observation of radio wave electromagnetic field on the ground. The achieved results in both the inversion problem and the wave direction finding are very promising and may support other works so far suggested the use of Bayesian methods in the inversion of ill-posed problems to benefit from the valuable information brought by the a priori knowledge.

This paper presents the performance prediction method on sentence recognition system which uses a finite state word automaton. When each word is uttered separately, the relationship between word recognition score and sentence recognition score can be approximated using the number of word sequences at a minimum distance from each sentence in the task. But it is not clear that how we get this number when the finite state word automaton is used as linguistic information. Therefore, we propose the algorithm to calculate this number in polynomial time. Then we carry out the prediction using this method and the simulation to compare with the prediction on the task of Japanese text editor commands. And it is shown that our method approximates the lower limit of sentence recognition score.

The relative intensity noise (RIN) spectra of DC driven 1.3 µm distributed feedback laser diodes under the influence of external reflections are measured for various currents and reflection lengths. The effective power reflectivities are 310-4-310-3. The enhanced noise is observed when the relaxation oscillation frequency coincides with the external cavity frequency. It is also observed that the RIN spectra with the near end reflections differ from those with the far end reflections. The degradation of the RIN spectra is analyzed with the rate equations numerically. A new reflection noise model, which includes the carrier density change induced by the reflections, is introduced. The near and far end reflections are characterized well by this model. Furthermore, it is found that the reflection induced noise effect can be described well by the far end reflection noise model even when the reflection length is as short as 1 m.

In this paper, we propose a new method to measure the 3D object shape without special purpose lighting based upon the Backprojection of Pixel Data.This method need not extract feature points such as edges from images at all and can measure not only the feature points but the whole object surface. It is simply done by project all pixel data back into the object space from each image. Actually, we first assign all pixel data of images into voxels in the object space, and evaluate the variance of assigned data for all voxels. This process is based on the idea that a point on the object surface gives the similar color information or gray level when it is observed from different view points. Then, two kinds of voting are executed as an enhancement process to eliminate the voxels containing the false points. We present experimental results under the circular constraint of camera movement and show the possibility of the proposed method.

A new approach is presented for recovering the 3-D shape from shading image. Photometric Stereo Method (PSM) is generally based on the direct illumination. PSM in this paper is modified with the indirect diffuse illumination method (IDIM), and then applied to hybrid reflectance model which consists of two components; the Lambertian reflectance and the specular reflectance. Under the hybrid reflectance model and the indirect diffuse illumination circumstances, the 3-D shape of objects can be recovered from the surface normal vector extracted from the surface roughness, the surface reflectance ratio, and the intensity value of a pixel. This method is rapid because of using the reference table, simplifies the restriction condition about the reflectance function in prior studies without any loss in performance, and can be applied to various types of surfaces by defining general reflectance function.

This paper presents Ternary Decision Diagrams which represent sets of products. This paper also presents manipulating methods for sum-of-products forms and ringsum-of-products forms using Ternary Decision Diagrams, and gives comparison results between Ternary Decision Diagrams and Binary Decision Diagrams.

Ultrasonic diffraction image of specimen informs its acoustic structure as X ray diffraction method for analysis of the crystal structure. This ultrasonic diffraction method has a feature that focused ultrasound beam is used and diffraction image is observed on focal plane. When the structure of specimen is perfectly periodic, its diffraction image produces symmetrical respect to origin, but the diffraction image of weak periodic structure such as living tissue has some asymmetricity. In this paper, the principle of ultrasonic diffraction method, and data processing for asymmetrical and scattered diffraction image caused by weak periodic structure are described. The results of diffraction image of plant tissue and animal tissue, and its discussion are also described. This method is expected to be useful in evaluation of acoustic structure such as living tissue and internal tissue of bone.

The aliasing probability was analyzed for MISRs when the error probability for each input was different. A closed form expression was derived by applying the complete weight distributions of linear codes over a Galois field and its dual codes. The aliasing probability for MISRs characterized by non-primitive polynomials was also analyzed. The inner product for binary representation of symbols was used instead of multiplication over a Galois field. The results show the perfect expression for analyzing the aliasing probability of MISRs.

A new method is proposed for realizing a flexible change detection which is free from the limitation that multitemporal images must have the same spectral bands whose center wavelength and bandwidth are identical. As spaceborne multispectral scanners are continuously improved for performance and new scanners do not necessarily have the same spaectral bands for observation, this limitation is a serious obstacle for detecting long term temporal change. The proposed method removes this limitation by using an image normalization technique based on multiple regression analysis. The method is successfully applied to actual remotely sensed multitemporal images.

In this study, an extraction method of failure sound signal which is strongly contaminated by noise is investigated by genetic algorithm and statistical tests of the frequency domain for the failure diagnosis of machinery. In order to check the extraction accuracy of the failure signal and obtain the optimum extraction of failure signal, the "existing probability Ps (t*k) of failure signal" and statistical information Iqp are defined as the standard indices for evaluation of the extraction results. It has been proven by practical field data and application of the inspection and diagnosis robot that the extraction method discussed in this paper is effective for detection of a failure and distinction of it's origin in the diagnosis of machinery.

This paper proposes a new method that can robustly recover 3D structure and 3D motion of 3D moving objects from a few multi-views. It recovers 3D feature points by obtaining intersections of back-projection lines which are connected from the camera's optical centers thorough projected feature points on the image planes corresponding to the different cameras. We show that our method needs only six views to suppress false 3D feature points in most cases by discussing the relation between the occurrence probability of false 3D feature points and the number of views. This discussion gives us a criterion to design the optimal multi-camera system for recovering 3D structure and 3D motion of 3D moving objects. An experimental multi-camera system is constructed to confirm the validity of our method. This system can take images from six different views at once and record motion image sequence from each view over a period of a few seconds. It is tested successfully on recovering the 3D structure of Vinus's plaster head and on recovering the 3D structure and 3D motion of a moving hand.

In VLSI or PCB layout, one often encounters a region that is either of rectilinear shape or can be approximated by a rectilinear region. Although many placement methods have been proposed, most of them are applicable only to rectangular regions. For these algorithms to be applied to a rectilinear region, two processing steps, region partitioning and rectangular region cell placement are necessary. Hence, the placement results are so far dependent on the locations of the regions partitioned and frequently become trapped in local minima. Recently, neural networks have been suggested as a new way to resolve the cell placement problem. This paper proposed a unified modeling method that uses a neural net model with additional calibration nodes to model rectilinear region cell placement. In this method, the ideal distance between cells is preserved to simultaneously minimize both the total wire length and the module overlap. Unlike traditional approaches, the proposed algorithm requires only a single processing step. Experiments have been conducted to verify the performance of the proposed algorithm. The total wire length obtained by our method is shorter than those generated by previous methods.

In this paper we describa a system for Off-line Recognition of Arabic characters and Numerals. This is based on expressing the machine printed Arabic alpha-numerical text in terms of strokes obtained by MCR (Minimum Covering Run) expression. The strokes are rendered meaningful by a labeling process. They are used to detect the baseline and to provide necessary features for recognition. The features selected proved to be effective to the extent that with simple right to left analysis we could achieve interesting results. The recognition is achieved by matching to reference prototypes designed for the 28 Arabic characters and 10 numerals. The recognition rate is 97%.

Mathematical morphology is inheriently suitable for range image processing because it can deal with the shape of a function in a natural and intuitive way. In this paper, a new approach to the extraction of the corner-edge-surface structure from 3D range images is proposed. Morphological operations are utilized for segmenting range images into smooth surface regions and high-variation surface regions, where the high-variation surface regions are further segmented into regions of edge type and regions of corner type. A new 3D feature, HV-skeleton, can be extracted for each high-variation surface region. The HV-skeletons can be thought of as the skeletons of high-variation surface regions and are useful for feature matching. The 3D features extracted by our approach are invariant to 3D translations and rotations, and can be utilized for higher-level vision tasks such as registration and recognition. Experimental results show that the new 3D feature extraction method works well for both simple geometric objects and complex shaped objects such as human faces.

With the fully depleted ultra-thin-film SOI CMOS, one important issue is controlling the threshold voltage (Vth) while maintaining high speed operation and low power consumption. To control the Vth, applying a bias voltage to the substrate is one of the most practical methods. We suggest a fully depleted ultra-thin-film SOI CMOS with a floating back gate, which is formed at the lower part of the channel field inside the substrate and stores electrons injected into it. This device can eliminate the necessity of an extra circuit or a separate power supply to apply a negative voltage. The silicon wafer direct bonding technique is used to construct this device. With the prototyped devices, we can successfully control the Vth for both the nMOSFET and pMOSFET at around 0.5 V by controlling the quantity of the electric charges injected into the floating back gate.