上浦 尚武

This paper proposes an on-line testing procedure for fuzzy controllers. A stuck-at-0 fault and a stuck-at-1 fault in the membership function are assumed as the fault models. In the proposed method, up to 7 extra functions are considered for each input variable in the antecedent part. The antecedent part and the consequent part can be tested in parallel. In the testing of the antecedent part, the sum of function values (degrees) derived based on the matching between the current measured input and the functions is observed. In the testing of the consequent part, the sum of the area of the output fuzzy sets, the sum of the degrees of applicability, and the sum of 7 degrees in the output fuzzy sets are observed. These procedures make it possible to handle simultaneously a fault in the antecedent part and a fault in the consequent part. Through comparisons with other methods, the proposed method is shown to be useful in terms of the number of faults that can be handled and the ease of application to multi-input controllers.

Clustering methods, such as k-means, Fuzzy C-Means (FCM), and others have been developed. However, they are only partitioning a database, so it is difficult to discover the reason why each cluster is formed. This paper proposes a method to discover the knowledge of how the clusters are derived. To select the center vector of each cluster, we employ an unsupervised clustering method based on Self-Organizing Map (SOM) without giving the number of clusters, We define the degree of contribution calculated from the weights of the neural network which learned the center vectors. We then describe the knowledge discovery method from the degrees. We applied our method to the artificial data and the clustering problem. The results show that the degree of contribution is an efficient indicator to represent the knowledge of how the clusters are formed.

In this paper we propose a fuzzy controller with the capability of compensating the influence of single stuck-at faults in membership functions. To cope with the faults in each antecedent part. the degree in a candidate regarded as a fault v function is exchanged for one of the following: 0. the degree in the next function to the candidate, and the difference between the constant and the degree in the next function to the candidate. To cope with the faults in the consequent part, several fuzzy variables are shifted. and then the fuzzy inference is executed with the membership functions representing shifted variables. Experimental results for a commercial controller show that the influence of ally single fault deviating the normal deterministic output of the controller is compensated completely.

In this paper we propose a learning algorithm to enhance the fault tolerance of feedforward neural networks (NNs for short) by manipulating the gradient of sigmoid activation function of the neuron. We assume stuck-at-0 and stuck-at-1 faults of the connection link. For the output layer, we employ the function with the relatively gentle gradient to enhance its fault tolerance. For enhancing the fault tolerance of hidden layer, we steepen the gradient of function after convergence. The experimental results for a character recognition problem show that our NN is superior in fault tolerance, learning cycles and learning time to other NNs trained with the algorithms employing fault injection, forcible weight limit and the calculation of relevance of each weight to the output error. Besides the gradient manipulation incorporated in our algorithm never spoils the generalization ability.

In this paper we propose the on-line multiple-fault-detection of fuzzy controllers. The membership functions in each antecedent part are divided into at most four groups. The faults in each group are detected by observing calculations concerning degrees of the functions in it. Our test forms two output fuzzy sets perfunction, and observes the calculations concerning areas of the sets. It has the advantage of being valid for the case where the number of overlapped functions is three or more.

In this paper we propose a fuzzy controller with the capability of compensating the influence of single stuck-at faults in membership functions. To cope with the faults in each antecedent part, the degree in a candidate regarded as a faulty function is exchanged for one of the following: O, the degree in the next function to the candidate, and the difference between the constant and the degree in the next function to the candidate. To cope with the faults in the consequent part, several fuzzy variables are shifted, and then the fuzzy inference is executed with the membership functions representing shifted variables. Experimental results for a commercial controller show that the influence of any single fault deviating the normal deterministic output of the controller is compensated completely. © 2001 IEEE.

In this paper we propose a MIN (Multistage Interconnection Network) whose performance in the faulty case degrades as gracefully as possible. We focus on a two-dilated baseline network as a sort of MIN. The link connection pattern in our MIN is determined so that all the available paths established between an input terminal and an output terminal via an identical input of a SE (Switching Element) in some stage will never pass through an identical SE in the next stage. Extra links are useful in improving the performance of the MIN and do not complicate the routing scheme. There is no difference between our MIN and other constructed from a baseline network with regard to numbers of links and cross points in all SEs. The theoretical computation and simulation-based study show that our MIN is superior to others in performance, especially in robustness against concentrated SE faults in an identical stage.

In this paper we propose feedforward neural networks (NNs for short) tolerating multiple-valued stuck-at faults of connection weights. To improve the fault tolerance against faults with small false absolute values, we employ the activation function with the relatively gentle gradient for the last layer, and steepen the gradient of the function in the intermediate layer. For faults with large false absolute values, the function working as filter inhibits their influence by setting products of inputs and faulty weights to allowable values. The experimental results show that our NN is superior in fault tolerance and learning time to other NNs employing approaches based on fault injection, forcible weight limit and so forth.

Volume visualization of cerebral blood vessels is highly significant for diagnosis of the cerebral diseases. It is because the automated segmentation of the blood vessels from an MR angiography (MRA) image is a knotty problem that there are few works on it. This paper proposes an automated method to segment the blood vessels from 3D time of flight (TOF) MRA volume data. The method consists of: (1) removing the background, (2) volume quantization by watershed segmentation, and (3) classification of primitives by using an artificial neural network (NN). In the proposed method, the NN classifies each primitive, which is a clump of voxels, by evaluating the intensity and the 3D shape. The method was applied to seven MRA data sets. The evaluation was done by comparing with the manual classification results. The average classification accuracy was 80.8%. The method also showed the volume visualizations using target maximum intensity projection (target MIP) and surface shaded display (SSD). The evaluation by a physician showed that unclear regions on the conventional image were clearly depicted on applying the method, and that the produced images were quite interesting for diagnosis of cerebral diseases such as aneurysm and encephaloma. The quantitative and qualitative evaluations showed that the method was appropriate for blood vessel segmentation.

This paper proposes feed forward neural networks (NNs) tolerating stuck-at faults of weights. To cope with faults having small false absolute values, the potential calculation of the neuron is modified, and the gradient of activation function is steepened. To cope with faults having large absolute values, the function working as filter sets products of inputs and faulty weights to allowable values. The experimental results show that the proposed NN is superior in fault tolerance, learning cycles and time to other NNs.

In this paper we propose a fault tolerant baseline network as a sort of MINs (multistage interconnection networks) and discuss its performance analysis. For our MIN with N input and N output terminals, switching elements in the first and n-th stages are duplicated where n = log2N. Four-input two-output switching elements and two-input four output ones employed in the second and (n-1)-th stages are useful in sharing loads efficiently on the first and n-th stages respectively. The comparison results show that the theoretical throughput of our MIN without faults and the performance of our MIN with faults are superior to those of previously known ELMIN though our MIN requires slightly more hardware overhead than ELMIN.

To enhance fault tolerance ability of the feedforward neural networks (NNs for short) implemented in hardware, we discuss the learning algorithm that converges without adding extra neurons and a large amount of extra learning time and cycles. Our algorithm modified from the standard backpropagation algorithm (SBPA for short) limits synaptic weights of neurons in range during learning phase. The upper and lower bounds of the weights are calculated according to the average and standard deviation of them. Then our algorithm reupdates any weight beyond the calculated range to the upper or lower bound. Since the above enables us to decrease the standard deviation of the weights, it is useful in enhancing fault tolerance. We apply NNs trained with other algorithms and our one to a character recognition problem. It is shown that our one is superior to other ones in reliability, extra learning time and/or extra learning cycles. Besides we clarify that our algorithm never degrades the generalization ability of NNs although it coerces the weights within the calculated range.

This paper proposes a method for extracting the human hippocampus based on multiscale structure matching scheme. Focusing on the feature that an overextraction occurs on anatomically specific place, the method detects the redundancy by comparing with given desired models. Since each of the desired models has information about locations of their redundant segments, the place of corresponding redundancy can be specified on the overextracted object. Then, subtle intensity difference around their connecting place is investigated to separate the hippocampus and redundancy. The matching process can proceed in parallel for various types of redundancy and individual variances. Qualitative evaluation of a physician shows that our method can detect the redundancies and extract hippocampus correctly.

This paper shows an application of fuzzy information granulation (fuzzy IC) to medical image segmentation. Fuzzy IG is to derive fuzzy granules from information. In the case of medical image segmentation, information and fuzzy granules correspond to an image taken from a medical scanner, and anatomical parts, namely region of interests (ROIs), respectively. The proposed method to granulate information is composed of volume quantization and fuzzy merging. Volume quantisation is to gather similar neighboring voxels. The generated quanta are selectively merged according to degrees for pre-defined fuzzy models that represent anatomical knowledge of medical images. The proposed method was applied to blood vessel extraction from three-dimensional time-of-Right (TOF) magnetic resonance angiography (MRA) images of the brain. The volume data studied in this work is composed of about 100 contiguous and volumetric MM images. According to the fuzzy IG concept, information correspond to the volume data, fuzzy granules corresponds to the blood vessels and fat. The qualitative evaluation by a physician was done for two- and three-dimensional images generated from the obtained blood vessels. The evaluation shows that the method can segment MRA volume data, and that fuzzy IG is applicable to, and suitable for medical image segmentation.

In this paper we propose a fault tolerant baseline network as a sort of MINs (multistage inter-connection networks) and discuss its performance analysis. For our MIN with N input and N output terminals, switching elements in the first and n-th stages are duplicated where n=log(2)N. Four-input two-output switching elements and two-input four output ones employed in the second and (n-1)-th stages are useful in sharing toads efficiently on the first and n-th stages respectively. The comparison results show that the theoretical throughout of our MIN without faults and the performance of our MIN with faults are superior to those of previously known ELMIN though our MIN requires slightly mole hardware overhead than ELMIN.

In this paper we propose concurrent compensation for fuzzy controllers. The concurrent fault location is executed by observing each sum of two degrees of adjacent membership functions. Instead of the faulty degree in the antecedent part, we employ either 0 or degree of next membership function to the faulty one at the easily calculated abscissa. To compensate the faults in the consequent part,,ve shift several fuzzy variables, and infer,vith the membership functions representing the variables after shifts. The amount of shifting the variables is determined systematically. Experimental results show that our method is valid for any non-redundant single stack-at fault both in each of antecedent parts and in the consequent part.

To enhance fault tolerance ability of the feedforward neural networks (NNs for short) implemented in hardware, we discuss the learning algorithm that converges without adding extra neurons and a large amount of extra learning time and cycles. Our algorithm modified from the standard backpropagation algorithm (SBPA for short) limits synaptic weights of neurons in range during learning phase. The upper and lower bounds of the weights are calculated according to the average and standard deviation of them. Then our algorithm reupdates any weight beyond the calculated range to the upper or lower bound. Since the above enables us to decrease the standard deviation of the weights, it is useful in enhancing fault tolerance. We apply NNs trained with other algorithms and our one to a character recognition problem. It is shown that our one is superior to other ones in reliability, extra learning time and/or extra learning cycles. Besides we clarify that our algorithm never degrades the generalization ability of NNs although it coerces the weights within the calculated range.

This paper proposes an architecture of a registration system for medical images. Image registration is the process of determining correspondence between all points in two images of the same scene, and is now widely used to medical images. In medical imaging, segmentation, registration and interpolation play primary roles. In those, registration is the most time consuming task because we must compare all voxel data and then evaluate the matching degree many times. Quantitative evaluation criterion of matching degree with multiple-valued coding of the image feature is proposed, and all architecture to save the processing time of the data comparison is described Finally, as a practical application, we described the summary of a registration of human brain MR volume data to diagnose brain disease.

In this paper we propose controllability and observability measures to guide the D-algorithm for multiple-valued logic circuits. The former is determined in one forward transversal of the circuit, and used in determining the line where the consistency operation should proceed The latter is determined in one backward traversal, and used in executing the D-drive at the fanout point. Our measures are computed by simple recursive formulas, and the time required for computing them is relatively short. The experimental results show that our measures are helpful in reducing the number of times for backtracking.

As a sort of MINs (Multistage Interconnection Networks), we propose the 2-dilated baseline network whose performance in the fault case degrades as gracefully as possible. All the available paths established between an input terminal and an output one via an identical input of SE (Switching Element) in some stage never pass through an identical SE in the next stage. The loads on SEs, therefore, are shared efficiently. Extra links added to enhance the performance never complicate the routing scheme. There is no difference between our MIN and other ones in hardware overhead. Besides our MIN is superior to other ones in performance, especially in robustness against concentrated SE faults in an identical stage.

We propose a learning algorithm to enhance the fault tolerance of feedforward neural networks (NNs for short) by manipulating the gradient of sigmoid activation function of the neuron. For the output layer, we employ the function with the relatively gentle gradient. For the hidden layer, we steepen the gradient of function after convergence. The experimental results show that our NNs are superior to NNs trained with other algorithms employing fault injection and the calculation of relevance of each weight to the output error in fault tolerance, learning cycles and time. Besides our gradient manipulation never spoils the generalization ability.

本論文では, ファジーコントローラに対するオンラインテストを提案する. 故障モデルには, メンバシップ関数の0故障及び1故障なる縮退型故障を仮定する. 本方法では, 1入力変数につき前件部に最大7個の検査用関数を付加し, 前件部と後件部のテストを並列に行う. 前件部検査では, 入力値と各関数とのマッチングで得られた関数値(ディグリー)の和を観測する. 後件部検査では, ファジー推論結果の面積, 適合度の和, 出力ファジー集合上のディグリー7個の和を観測する. これにより, 前件部と後件部に1個ずつ同時に生じた故障も検出対象とできる. また, 他方法との比較により, 本方法は対応可能な故障の数及び多入力コントローラへの適用容易性の観点から有利なことを示す.

本文では, トリー状結合プロセッサの再構成法及びレイアウト法を提案する. 本方法と従来法とを比較し, 最大配線長の観点から本方法が有利であることを示す.

本文では, 多段相互接続網(Multistage Interconnection Network:MIN)の一つであるべースライン網の耐故障設計法とその性能解析について述べる. 本方法では, スイッチ段数nのべースライン網に対し, 第1段と第n段のスイッチを多重化するとともに, 第2段と第(n-1)段にそれぞれ4入力2出力および2入力4出力のスイッチを用いる. これにより, 任意のターミナル間に経路が複数確保されるので, 耐故障性および性能が向上する. 本方法と従来法のMINを比較した結果, 本方法は最大でも1.34T倍程度となる比較的少ないハードウェア増加で故障時および非故障時のスループットを向上できることが示される.

本文では, 多段相互接続網(Multistage Interconnection Network:MIN)の一つであるべースライン網の耐故障設計法とその性能解析について述べる. 本方法では, スイッチ段数nのべースライン網に対し, 第1段と第n段のスイッチを多重化するとともに, 第2段と第(n-1)段にそれぞれ4入力2出力および2入力4出力のスイッチを用いる. これにより, 任意のターミナル間に経路が複数確保されるので, 耐故障性および性能が向上する. 本方法と従来法のMINを比較した結果, 本方法は最大でも1.34T倍程度となる比較的少ないハードウェア増加で故障時および非故障時のスループットを向上できることが示される.

本文では, 多段相互接続網(Multistage Interconnection Network:MIN)の一つであるべースライン網の耐故障設計法とその性能解析について述べる. 本方法では, スイッチ段数nのべースライン網に対し, 第1段と第n段のスイッチを多重化するとともに, 第2段と第(n-1)段にそれぞれ4入力2出力および2入力4出力のスイッチを用いる. これにより, 任意のターミナル間に経路が複数確保されるので, 耐故障性および性能が向上する. 本方法と従来法のMINを比較した結果, 本方法は最大でも1.34T倍程度となる比較的少ないハードウェア増加で故障時および非故障時のスループットを向上できることが示される.

Clustering is primarily used to uncover the true underlying structure of a given data set. Most algorithms for fuzzy clustering often depend on initial guesses of the cluster centers and assumptions made as to the number of subgroups presents in the data. In this paper, we propose a method for fuzzy clustering without initial guesses on cluster number in the data set. Our method assumes that clusters will have the normal distribution. Our method can automatically estimates the cluster number and achieve the clustering according to the number, and it uses structured Genetic Algorithm (sGA) with graph structured chromosome.

This paper describes the benefit of utilizing the unary function generators in a multiple-valued Programmable Logic Array (PLA). We will clarify the most suitable PLA structure in terms of the array size. The multiple-valued PLA considered here has a structure with two types of function generators (literal and unary function generators), a first-level array and a second-level array. On investigating the effectiveness to reduce the array size, we can pick up four form PLAs: MAX-of-TPRODUCT form, MIN-of-TSUM form, TSUM-of-TPRODUCT form and TPRODUCT-of-TSUM form PLAs among possible eight form PLAs constructing from the MAX, MIN, TSUM and TPRODUCT operators. The upper bound of the array sizes with v UGs is derived as ([log2p]pv + p(n - v) + 1) pn-1to realize any n-variable p-valued function. Next, experiments to derive the smallest array sizes are done for 10000 randomly generated functions and 21 arithmetic functions. These results conclude that MAX-of-TPRODUCT form PLA is the most useful in reducing the array size among the four form PLAs.

In this paper, we Proposes a method of reconfiguring 2D meshes embedded in hypercubes. Our reconfiguration for link failures consists of two stages. The first stage assigns the d dimensions of the hypercubes to two directions with respect to rows and columns in the mesh, so that the number of disconnected pairs with adjacent rows and columns becomes smaller. The second stage re-establishes the mesh communication by assigning the Cartesian product of two Gray code sequences, which represent the order of healthy rows and columns to every node. We introduce graphs with edges corresponding to connections between rows or columns, and search the sequences. Then either an approach based on the depth-first search or one based on a genetic algorithm is applied to the graphs. The approach is made valid for node failures by regarding a faulty node as a node without healthy links. Simulation results show that our method reconfigures embedded meshes efficiently with relatively short computation time.

This paper proposes a thresholding based segmentation method aided by Kleene Algebra. For a given image including some regions of interest (ROIs for short) with the coherent intensity level assume that we can segment each ROI on applying thresholding technique. Three segmented states are then derived for every ROI: Shortage denoted by logic value 0 Correct denoted by 1 and Excess denoted by 2. The segmented states for every ROI in the image can be then expressed on a ternary logic system. Our goal is then set to find "Correct (1)" state for every ROI. First unate function which is a model of Kleene Algebra based procedure is proposed. However this method is not complete for some cases that is correctly segmented ratio is about 70% for three and four ROI segmentation. For the failed cases Brzozowski operations which are defined on De Morgan algebra can accommodate to completely find all "Correct" states. Finally we apply these procedures to segmentation problems of a human brain MR image and a foot CT image. As the result we can find all "1" states for the ROIs i.e. we can correctly segment the ROIs.

In this paper we discuss the reconfiguration of two-dimensional meshes embedded in hypercubes with link and/or node failures. First, we assume that only the link failures may occur. Our method consists of two stages. The first stage assigns dimensions of hypercube to two directions of mesh so that the losses of rows or columns would be as small as possible. The second stage establishes the mesh communication by assigning the Cartesian product of two Gray code sequences to every node. We generate these sequences with a depth-first search or generic algorithm. This method can be applied to node failures by regarding a faulty node as a node whose links are entirely faulty. Our simulation results show that our method can reconfigure large meshes with short computation time.

In this paper we discuss the reconfiguration of two-dimensional meshes embedded in hypercubes with link and/or node failures. First, we assume that only the link failures may occur. Our method consists of two stages. The first stage assigns dimensions of hypercube to two directions of mesh so that the losses of rows or columns would be as small as possible. The second stage establishes the mesh communication by assigning the Cartesian product of two Gray code sequences to every node. We generate these sequences with a depth-first search or generic algorithm. This method can be applied to node failures by regarding a faulty node as a node whose links are entirely faulty. Our simulation results show that our method can reconfigure large meshes with short computation time.

本文では, ファジィインフォメーショングラニュレーション(fuzzy Information Granulation; fuzzy IG)による, 頭部MR画像からの脳領域抽出の為の濃度しきい値発見法を提案する.Fuzzy IGとは, Zadehが提唱する新しい概念である.ファジィ情報(fuzzy information)は, 幾つかのファジィグラニュール(fuzzy granule)で構成され, fuzzy IGによってfuzzy informationからfuzzy granuleが引き出される.本手法では, fuzzy informationを濃度ヒストグラムとして, fuzzy granuleを脳を構成する白質部, 灰白質部, 脳脊髄液のヒストグラム上のピークとする.Fuzzy IGは, 医師の知識をモデル化したヒストグラムと, 画像の濃度ヒストグラムとのファジィマッチングによって行う.得られたしきい値を用いて自動抽出した脳体積と, 手動で得られた体積とを, 50例のMR画像に対して比較した結果, 平均誤差率は2.3%であった.

<p>本文ではファジィ推論に基づく脳部位の自動分割法を提案する.分割対象とする部位は右脳, 左脳, 小脳および脳幹である.本方法では各部位について位置, 形状および濃度の知識を与え, ボクセルが部位間境界に所属する度合いを推論する.推論は2段階で行われる.まず, 位置に関する知識で全体を大きく4つの領域に分割する.つぎに, 形状および濃度の知識が厳密な境界所属度を決定していく.得られた推論結果はリージョングローイングの過程で評価され, 各部位が自動的に抽出される.本方法を36人の脳MR画像に適用した結果, 医師による分割に対する平均誤差率は2.5%であった.</p>

This paper proposes a segmentation method based on Kleene Algebra. for an input image including some regions of interests (ROIs for short), consider three segmented states: Shortage, correct, Excess for the target region on applying segmentation method based on standard intensity thresholding. For the target image, we do thresholding to each of ROIs, then to derive all "Correct" for ROIs, unate function (one model of Kleene Algebra) based approach proposes to find all "Correct" states. However, the method is not complete for some cases, that is, correctly segmented ratio is about 70% for three and four ROI segmentation. For the failed cases, it is proved that Brzozowski operations are provided to completely find all "Correct" states. The experimental results on a humanitarian MR image and a foot CT image show that our method can correctly segment the ROI.

This paper presents a feature extraction model 'MAGNET' to find the deepest point of branched sulcus. Our model demonstrates magnetic principle and consists of four types of ideal magnetic poles: an N-pole and three S-poIes. According to attractive or repulsive Coulomb forces between their poles, one of the S-poles is pushed to the deepest point of the sulcus. First, we explain our model on the simple sulcus model. Second, we apply it to the sulcus with implicit branches. Our model can detect the target points in every branch. Then an example to realize the model on a synthetic image is introduced. We apply our model to human brain MR images and human foot CT images. Experimental results on human brain MR images show that our method enable us to successfully detect the points.

In this paper, we propose concurrent tests of fuzzy controllers. We assume multiple stuck-at faults in membership functions. For tests of antecedent membership functions, we observe the sum of degrees by matching a current measured input with each function and that of degrees at several easily calculated co-ordinates, For tests of consequent ones, we absente the sum of their maximum degrees, that of degrees of applicability and that of degrees at several co-ordinates an the union of output fuzzy sets. We can find all the multiple faults except redundant ones. Our tests are superior to other tests in the number of additional membership functions and multiple fault detection.

This paper proposes a scheme of image granulation by fuzzy inference technique. For a region of interests(ROI) in a medical image, we describe knowledge needed to granulate the ROI, for example, knowledge of intensity, knowledge of location and so on. Generally, we cannot granulate the ROI without employing the whole of the knowledge. Fuzzy inference rules of the derived knowledge can accommodate the granulation. After the inference results are compiled to a total degree, a resultant data is obtained. Clustering or region growing technique is used to granulate the ROI. The experimental results on human brain MR images and human foot CT images show that our method can precisely granulate the ROI.

In this paper, we discuss a reconfigurable tree-connected multiprocessor system and its arraylike layout. Each level in our tree consists of several blocks with PEs. The reconfiguration is executed for each block by shifting PEs to the right. it is valid if the number of faulty PEs in each block is less than or equal to that of spare ones in it. We introduce a 7x7 square module with a five-level tree to simplify the array like layout. The system with sir or more levels is constructed easily by arranging several modules regularly. The comparison with other trees layoutable in the planar arrays shows that our tree is superior to others irt maximum interconnection length.

One of the most serious problems in medical science is an increment of Alzheimer's disease. It is known that the patient's brain atrophy is a result of neural cell loss. It is useful for the diagnosis of Alzheimer's disease to measure the volumes of the brain portions and to display them. We can obtain anatomical information from 2D slice images produced by MRI. We propose a computer-aided system for the diagnosis of Alzheimer's disease. The system consists of: (1) extraction of the portions from MRI data, (2) measurement of the volumes of the portions and then displaying them, and (3) user interface for a medical doctor. In this paper we describe procedures for the above. For extraction of the brain portions we propose the method based on standard region growing algorithm and the method of figure decomposition using the distance value. Comparison of the volumes of our extracted portions with volumes manually measured by a physician shows that the error rate, on the average, is 1.74% for 48 MRI data. We also discuss the 3D display, the measuring range, and the construction of the user interface for a physician. © 1997 Scripta Technica, Inc.

MR脳画像同士の位置あわせは, 同じ脳疾患を持つ複数の患者の脳画像を重ねあわせることによって, 濃度値などの情報から共通する部位の異常を視覚的に捉えることが目的である. しかし, MR脳画像は大局的に見ると一様に見える部分であっても, 局所的には個々により異なる部分が多く, このことがMR脳画像同士の位置あわせを困難としている. 従来, 画像同士の位置あわせの方法がいくつか紹介されている. それらは, 画像の濃度値による相関関係による方法, 輪郭の曲率や変極点の位置情報による方法などである. 相関関係による方法のような対象画像の濃度値をもとに行われる方法は, 濃度値情報を取得するというMR脳画像を位置あわせする目的から使用することができない. また, これらの方法は対象画像の局所的な一致度に基づいて全体の位置あわせを行うものであり, 局所的に多くの部分が異なる脳画像の位置あわせに適用するのは困難である. 本文では, ラバー (ゴム) をモデル化した物理モデルを提案し, このラバーモデルを用いてMR脳画像の位置あわせを行う. ラバーモデルとは, 平面状のゴムを想定し, このゴムに歪みが加わったときのゴム全体の力学的な振る舞いをモデル化したものである. ラバーモデルでは局所的な変位が画像全体に影響を及ぼす構造をしており, 全体として協調しながら画像全体が移動するため, 大局的な位置あわせが可能である.

近年、聴覚障害者と健聴者とのコミュニケーションの支援を目的とし、手話を日本語へ翻訳する研究が注目されている。我々は、日本語の五十音に相当する手話の指文字のうち、表現に動きを用いない 41 文字を対象として画像処理を用いた聴覚障害者のためのインタフェースについて検討を行ってきた[1]。今回は手形状の伸展指特徴による指文字の分類手法を示し、手と照明との位置関係による画像不良に対処するために用いたグローブの有用性について検討した結果を報告する。

現在, ニューラルネットワークは画像処理やシステム制御などのさまざまな分野に応用され, その有効性が示されている. しかしこれらの研究のほとんどが, 与えた入力に対する出力値や学習に対する反応のみに注目しており, ネットワークの内の各ユニットがどのような機能を実現しているかには, あまり注意が払われていない. 与えられた入出力パターンを学習したネットワークは, 設計者の望む機能を実現しており, 入出力パターンから何らかの知識を獲得しているはずである. しかし, ネットワークが獲得した知識は複雑で, 人間に理解できる形で導出することは非常に困難である. 本研究では, 加重和のかわりにmin演算やmax演算等を用いることで, ニューロンモデルと知識の対応を明確にし, ネットワークから知識を導出する. 同時に, ネットワークの枝刈りを行い中間層ユニットを削減することで, 知識の最適化を行う. この概念を論理回路の設計に適用する.

FPGA (Field Programable Gate Array) を設計する際, 与えられた多変数関数を数個の部分関数に分解する手続きである関数分解は, 回路コスト削除のための必要不可欠な手法である. これまでは, 真理値表を用いた関数分解法が多く用いられてきたが, これは, 複数段の回路構成には適用が困難だった. 本文では, 多段回路設計効率化のために, Multiple-valued Decision Diagram (MDD) 使用による関数分解法を提案する. MDDは実用的な多値論理関数を計算機内部で効率よく表現できるグラフとして知られており, 関数分解にも有効である期待できる. 本方法の評価は, 多値算術関数に対する適用結果を基に行う.

近年, MRやCT装置などの医療機械の発展によって, 人体内部が簡単に撮影できるようになった. また, 最近のコンピュータの高速化, メモリの大容量化によって, それらの情報を解析・可視化し, 評価することが容易になってきた. 一般的には, 興味ある部位をその医療画像から抽出し, それを評価・可視化・解析することが必要である. その際, 抽出目的部位の濃度幅が異なるため単純なしきい値処理では抽出できないのが現状である. そこで, 本稿では, 複数存在する抽出目的部位毎にしきい値処理を行って得られた画像群を, クリーネ代数を用いて統合する方法を示す. 更に, クリーネ代数を用いても解が得られない場合に対して, Brzozowskiの演算を用いることで解が得られることを示す.

近年, VLSI(Very Large Scale Integration)技術の進歩とともに, 1枚のウェーハ上に超並列マルチプロセッサを搭載するWSIの(Wafer Scale Integration)が注目されている.本文では,超並列マルチプロセッサの一つである格子結合型ネットワークシステムの利点とトリー状結合型ネットワークシステムの利点の両方を有するネットワークシステムを提案するとともに,その欠陥救済法も明らかにする.また,システムの信頼性を評価するために,PE(Processing Element)の信頼度を変数とする信頼度関数を定義する.そして,従来の格子結合型ネットワークシステムと本システムとを信頼度関数に関して比較し,本システムの有効性について検討する.

本文では, ニューラルネットワークを用いたパターン発生のモデルについて報告する.このモデルの目的は, 1つのパターンから多数のパターンを発生させるメカニズムの実現である.ここでは, その実現に対して, 1.十分な中間層をもったニューラルネットワークに対して、ファジィ論理上で表現したパターン発生の核となるパターンをバックプロパゲーション法を用いて学習させる.2.そのニューラルネットの規模を学習とユニットの削除を繰り返すことにより最小化する.3.ユニットを新たに付加し、そのパラメータを乱数を用いて決定したニューラルネットに最初の核となるパターンを入力することで類似パターンを発生させる.以上の処理を有するモデルを提案する.本文ではこのモデルを単調論理関数の発生, 意思決定および画像パターンの発生に適用し, それぞれについて良好な結果が得られたことを示す.

A universal literal is a single-variable function and has an ability to manipulate more information than a set literal. The array size therefore could be eliminated by using universal literal generators (ULGs for short) in programmable logic arrays (PLAs), compared to PLAs with set literals. This paper discusses what operator is the most suitable in the term of eliminating the array size. We find four solutions as the good operator structures to eliminate the array size. A speculation of the upper bound of the array sizes is shown. Experiments are also done for randomly generated functions and some arithmetic functions. The experimental results show that the MAX-of-TPRODUCT form PLAs require the smallest array size.

This paper proposes an approach of fuzzy logic to 3D MR image segmentation. We show a fuzzy knowledge representation method to represent the knowledge needed to segment the target portions, and apply our method to 3D MR human brain image segmentation. In it we consider position knowledge, boundary surface knowledge and intensity knowledge. They are expressed by fuzzy if-then rules and compiled to a total degree as the measure of segmentation. The degree is evaluated in region growing technique and which segments the whole brain region into the left cerebral hemisphere, the right cerebral hemisphere, the cerebellum and the brain stem. The experimental result on 36 MR voxel data shows that our method extracted the portions precisely.

This paper describes an approach to minimize multiple-valued logic expressions by genetic algorithms. We encode a multiple-valued logic expression as a ''chromosome'' whose length allows to change and corresponds to the number of implicants of the expression. Our fitness function evaluates the following three items. 1. How may outputs does the logic expression represent correctly? 2. How many implicants does the logic expression require? 3. How many connections does the logic expression require? Our method employs the fitness function and minimizes sum-of-products expressions, where sum refers to TSUM or MAX and product refers to MIN of set literals or window literals. The simulation results show that our method derives good results for some arithmetic functions and intends to avoid the local minimal solution, compared to neural-computing-based method.