上浦 尚武

We propose rarity-oriented retrieval methods for serendipity using two approaches. We define rare information as relevant and atypical information. We propose two approaches. In the first approach, we use social bookmark data. We introduce tag estimation to our previous work. The second approach is based on word co-occurrence in a dataset. In both approaches, we use conditional probabilities to express relevancy and atypicality. In experiments, we compared our methods with the relevance-oriented method, the diversity-oriented method, and another rarity-oriented method. Our methods using word co-occurrence obtained better nDCG scores than the other methods.

Magnetic resonance (MR) images are widely used to diagnose cerebral diseases. The diseases may deform the brain shape, and the deformed region differs among types of diseases. To evaluate the brain shape deformation, MR image registration (IR) has been used. There are some IR methods for brain MR images but they mainly use MR signal based likelihood. We cannot directly apply methods for adult brain to neonatal brain because there are large differences in MR signal distribution and brain shape. This paper focuses on neonatal brain MR images, and introduces a sulcus extraction method using Hessian matrix based on a feature called sulcal-distribution index (SDI). SDI is calculated from MR signal on the cerebral surface. Next, this paper proposes an iterative closest point (ICP) based brain shape registration method using the extracted sulci. The proposed method will be effective for neonatal brain in which the accurate delineation of cerebral surface is difficult because the method evaluates the correspondence of cerebral sulci distribution. Results in seven neonates (modified age was between 3 weeks and 2 years) showed that the method registered one brain with the other brain successfully.

In this paper, a simple personal identification system is presented, using data associated with corneal thickness measured by OCT (Optical Coherence Tomography). The proposed method divides the cornea into thirty two fan-shaped segments, each of which has the same area, using thirty two radiuses. It generates the thirty-two-dimensional (or sixty-four-dimensional) vector for some test subject as the registered data at some time point, and adds it to a set. The data consists of element values equal to minimum values and/or maximum values on the above radiuses. When the test subject takes medical practice, the proposed method generates thirty-two-dimensional (or sixty-four-dimensional) vector for the corresponding test subject as the collation data, and calculates the Euclidean distance between the collation data with each of the vectors in the set. It judges the test subject corresponding to the given collation data to be that of the registered data with the shortest distance to the given data. Experimental results establish that the proposed method achieves one hundred percentage as the identification rate on assumption that the number of test subjects is 30, when element values of the registered data and collation data are associated with pachymetry.

© 2015. People are posting huge amounts of varied information on the Web as the popularity of social media continues to increase. The sentiment of a tweet posted on Twitter can reveal valuable information on the reputation of various targets both on the Web and in the real world. We propose a method to classify tweet sentiments by machine learning. In most cases, machine learning requires a significant amount of manually labeled data. Our method is different in that we use social bookmark data as training data for classifying tweets with URLs. In social bookmarks, comments are written using casual expressions, similar to tweets. Since tags in social bookmarks partly represent sentiment, they can be used as supervisory signals for learning. The proposed method moves beyond the basic "positive"/"negative" classification to classify impressions as "useful", "funny", "negative", and "other".

In this paper, a method of analyzing relationships between items in specific health examination data is presented to cope with lifestyle-related diseases. The proposed method uses self-organizing maps, and focuses on twelve items such as hemoglobin A1c (HbA1c), glutamic-oxaloacetic transaminase (GOT), glutamic-pyruvic transaminase (GPT), gamma-glutamyl transpeptidase (gamma-GPT), and triglyceride (TG). The proposed method picks up the data from the examination dataset according to the standard specified by some item values. The training data are then generated by calculating the difference between item values associated with successive two years and normalizing the values of this calculation. The proposed method labels neurons in the map by using item values of training data as parameters, and examine the relationships between items in the examination data by observing clusters formed in the map. Experimental results reveal the relationships among HbA1c, GOT, GPT, gamma-GTP and TG both in the unfavorable case of HbA1c deteriorating and in the favorable case of HbA1c being improved.

Cerebral atrophy treated as one of the common feature of many diseases that affect the brain. In general, atrophy means that all of the brain has shrunk or it can be regional, affecting a limited area of the brain which ends up resulting in neural decrease related to functions that area of brain controls. Detection of early brain atrophy can help physicians to detect the disease at curable stage. In this paper brain atrophy with some given landmark positions has been evaluated using dimensionality reduction methods. A comparative study has been done between principal component analysis and manifold learning using Laplacian eigenmaps to quantify brain atrophy. In addition, a novel method has been proposed with combination of PCA and Manifold learning which evaluates brain atrophy with corresponding age groups. Selection of principal component scores to optimize manifold learning parameters added effective feature to the findings. The method has been applied to open database (IXI database). We applied principal component analysis to deformation maps derived from MR images of 250 normal subjects. After sampling, 42 subjects were taken whose principal component scores were used to discriminate between older subject and younger subject. We found a significant regional pattern of atrophy between distance of Anterior Commissure, Posterior Commissure, Anterior Commissure to both frontal lobe, Posterior Commissure to both frontal lobe with corresponding age. After going through T-test principal component analysis showed the best value of significant difference (p<0.0036) over the Manifold learning (p<0.4095). The proposed method outperformed both the dimensionality reduction method with a score of (p<0.0030). Our findings indicates that multivariate network analysis of deformation maps detects typical feature of atrophy and provides a powerful tool to predict brain atrophy with age.

In this paper, a system of personal identification using data associated with corneal thickness is presented for ophthalmological patients. The data are measured by OCT (Optical Coherence Tomography). The proposed method equally divides the cornea into thirty-two fan-shaped segments, using thirty-two radiuses. It generates a couple of thirty-two-dimensional vectors (or a sixty-four-dimensional vector) for some test subject as the registered data, and adds it to a set. Each of the data consists of element values associated with minimum values and/or maximum values on the above radiuses. When the test subject takes medical practice, the proposed method generates two thirty-two-dimensional vectors (or a sixty-four-dimensional vector) for the corresponding test subject as the collation data. Then, the proposed method calculates the distance between the collation data with each of the vectors in the set. It judges the test subject corresponding to the given collation data to be that of the registered data with the shortest distance to the given data. Experimental results establish that the proposed method achieves 100 percent as the identification rate on assumption that the number of subjects is at most 45.

To prevent lifestyle diseases, this paper studies disease prediction using periodical health checkup data, daily monitoring to maintain healthy condition, and early life disease detection with medical imaging. To analyse periodical health checkup data, three approaches are introduced. The first approach is based on fuzzy set. It converts all attributes of health checkup data into fuzzy degrees by defining fuzzy membership functions. It enables us to manipulate all attributes in the same scale. The second approach analyses relationships between attributes of specific health examination data to cope with lifestyle diseases. It uses self-organizing maps, and clarifies the relationships among hemoglobin A1c (HbA1c), glutamic-oxaloacetic transaminase, glutamic-pyruvic transaminase, gamma-glutamyl transpeptidase, and triglyceride. The third approach predicts HbA1c fluctuations using decision tree. If we can predict the fluctuation, we can extract knowledge about what element will trigger developing diabetes. Through our examination, BMI will be the largest influencer about HbA1c fluctuations. Daily understanding of own condition is the first step of maintaining our health. A MEMS-based small and flexible monitoring device has been developed by the ERATO Maenaka human-sensing fusion project. We propose a condition estimation method using the monitoring device and FNN-based condition estimation. Experimental results show that it is a promising method for condition understanding. Cerebral vascular disease is one of major lifestyle diseases, and is caused by cerebral aneurysms. To predict the diseases, we should analyse cerebral arteries and aneurysms using magnetic resonance angiography images. This paper introduces an automated analysis method for early detection of aneurysms.

In this paper, we present a method of selecting intraocular lens (IOL) power formulas for cataract patients. The method is based on support vector machines (SVM's) and genetic algorithm (GA). We assume that each of patients' data belongs to one of the classes named by three power formulas. We therefore consider the formula selection to be the issue of classifying patients' data. Since three formulas have variables associated with axial length and corneal refractive power, we always employ them as elements in data. If there are elements probably useful in classifying the data, they are found by GA. We construct the final discrimination model by SVM learning using data with the above elements. The model consists of three coordinate spaces, each of which has two regions corresponding to two formulas. A space provides a potential solution. We finally take a majority of the potential solutions, and determine the formula suitable to some patient specified by it. We show that the proposed method achieves the most favorable percentage of concordance for the classification, if the two-point crossover technique is adopted in GA.

In this paper, a system of personal identification using data associated with corneal thickness is presented for ophthalmological patients. The data are measured by OCT (Optical Coherence Tomography). The proposed method equally divides the cornea into thirty-two fan-shaped segments, using thirty-two radiuses. It generates a couple of thirty-two-dimensional vectors (or a sixty-four-dimensional vector) for some test subject as the registered data, and adds it to a set. Each of the data consists of element values associated with minimum values and/or maximum values on the above radiuses. When the test subject takes medical practice, the proposed method generates two thirty-two-dimensional vectors (or a sixty-four-dimensional vector) for the corresponding test subject as the collation data. Then, the proposed method calculates the distance between the collation data with each of the vectors in the set. It judges the test subject corresponding to the given collation data to be that of the registered data with the shortest distance to the given data. Experimental results establish that the proposed method achieves 100 percent as the identification rate on assumption that the number of subjects is at most 45.

In this paper, we present a method of selecting intraocular lens (IOL) power formulas for cataract patients. The method is based on support vector machines (SVM's) and genetic algorithm (GA). We assume that each of patients' data belongs to one of the classes named by three power formulas. We therefore consider the formula selection to be the issue of classifying patients' data. Since three formulas have variables associated with axial length and corneal refractive power, we always employ them as elements in data. If there are elements probably useful in classifying the data, they are found by GA. We construct the final discrimination model by SVM learning using data with the above elements. The model consists of three coordinate spaces, each of which has two regions corresponding to two formulas. A space provides a potential solution. We finally take a majority of the potential solutions, and determine the formula suitable to some patient specified by it. We show that the proposed method achieves the most favorable percentage of concordance for the classification, if the two-point crossover technique is adopted in GA.

To prevent lifestyle diseases, this paper studies disease prediction using periodical health checkup data, daily monitoring to maintain healthy condition, and early life disease detection with medical imaging. To analyse periodical health checkup data, three approaches are introduced. The first approach is based on fuzzy set. It converts all attributes of health checkup data into fuzzy degrees by defining fuzzy membership functions. It enables us to manipulate all attributes in the same scale. The second approach analyses relationships between attributes of specific health examination data to cope with lifestyle diseases. It uses self-organizing maps, and clarifies the relationships among hemoglobin A1c (HbA1c), glutamic-oxaloacetic transaminase, glutamic-pyruvic transaminase, gamma-glutamyl transpeptidase, and triglyceride. The third approach predicts HbA1c fluctuations using decision tree. If we can predict the fluctuation, we can extract knowledge about what element will trigger developing diabetes. Through our examination, BMI will be the largest influencer about HbA1c fluctuations. Daily understanding of own condition is the first step of maintaining our health. A MEMS-based small and flexible monitoring device has been developed by the ERATO Maenaka human-sensing fusion project. We propose a condition estimation method using the monitoring device and FNN-based condition estimation. Experimental results show that it is a promising method for condition understanding. Cerebral vascular disease is one of major lifestyle diseases, and is caused by cerebral aneurysms. To predict the diseases, we should analyse cerebral arteries and aneurysms using magnetic resonance angiography images. This paper introduces an automated analysis method for early detection of aneurysms.

Magnetic resonance (MR) images are widely used to diagnose cerebral diseases. The diseases may deform the brain shape, and the deformed region differs among types of diseases. To evaluate the brain shape deformation, MR image registration (IR) has been used. There are some IR methods for brain MR images but they mainly use MR signal based likelihood. We cannot directly apply methods for adult brain to neonatal brain because there are large differences in MR signal distribution and brain shape. This paper focuses on neonatal brain MR images, and introduces a sulcus extraction method using Hessian matrix based on a feature called sulcal-distribution index (SDI). SDI is calculated from MR signal on the cerebral surface. Next, this paper proposes an iterative closest point (ICP) based brain shape registration method using the extracted sulci. The proposed method will be effective for neonatal brain in which the accurate delineation of cerebral surface is difficult because the method evaluates the correspondence of cerebral sulci distribution. Results in seven neonates (modified age was between 3 weeks and 2 years) showed that the method registered one brain with the other brain successfully.

頭部MRA画像からの脳血管領域抽出は,動脈瘤検出や,脳血管構造理解のため,必要不可欠の処理である.MRA画像はMRI装置での撮影であるため,装置間,画像間でコントラストの違いが大きく,全自動化処理の妨げとなっている.そこで,本研究では,ファジィ連結度を用いた脳血管領域の自動抽出法を提案し,同手法において解析パラメータの全自動設定法を示す.すなわち,本手法では,完全に脳血管領域処理を自動化,パイプライン化することができる.実験では,磁気強度の異なる複数装置で撮影された,様々な症状を有する被験者に本手法を適用した.本手法は,磁気強度の違い,動脈瘤の有無,糖尿病,高血圧,高脂血症などの他疾患の有無に関わらず,良好に脳血管領域を自動抽出できた.

脳機能解析や VBM(voxel-based-morphometry)などにおいてMR画像を用いた個人脳間の画像位置合わせが用いられている．従来法では主に画像位置合わせの尤度を，脳全体の信号値の一致度とし，脳溝の一致を評価していないため，異なる脳回に位置合わせされる危険性がある．また，新生児脳はMR信号特徴が異なり，脳溝が狭いため，成人を対象とする従来法の適用は困難である．本文では，脳表近傍のMR信号値から算出される脳溝特徴分布(SDI; sulcal-distribution index)を用いたFlatteningにより，真球上でSDI相互情報量を最大化する3次元非剛体変形を行い，脳溝の一致度を尤度とした脳形状位置合わせを可能とする．また，変形時に緩和項としてバネモデルを使用することで真球上の制御点密度の偏りを減少させる．提案法を修正齢3から5週間の新生児3名のMR画像に適用した結果，緩和項を用いることで位置合わせ精度の向上が確認できた．

Brain region segmentation in neonatal magnetic resonance (MR) images is an essential task for computer-aided diagnosis of neonatal brain disorders using MR images. We have proposed a neonatal brain segmentation method using a fuzzy object model (FOM), which represents a prior knowledge of brain shape and location. The FOM is constructed from multiple neonatal brain MR images whose revised age was between 0 and 4 weeks. The method segmented the brain region with a good accuracy for subjects whose age matches of the training data set. To enhance the method, we need multiple FOMs for each age. The other solution is to develop a growable model. This paper introduces 4-D FOM and applies it to neonatal brain segmentation. This paper introduces a neonatal brain segmentation method using 4-D FOM. The proposed method consists of three components. The first part proposes a method for estimating the brain development progress, called growth index in this study, from MR images based on Manifold learning. The second part shows a procedure for generating 4-D FOM using the estimated growth index. The third part is to segment brain region based on fuzzy-connectedness image segmentation using 4-D FOM. The proposed method was applied to 16 neonatal subjects. The results show that 4-D FOM is superior to stable 3-D FOM for segmenting neonatal brain region from MR images.

Total hip arthroplasty (THA) is an orthopaedic surgery which replaces the damaged hip joint with implants. The acetabular cup is implanted to the acetabulum. Some studies show that the outcome of THA is strongly correlated to the orientation of the acetabular cup. This paper proposes a fully automated method for measuring the orientation using multidetector-row computed tomography (MDCT) images. The method defines the pelvic anatomical coordinate system using anterior pelvic plane (APP), and measures angles between the cup implanting axis and the pelvic anatomical coordinate axis. The angles are inclination angle and anteversion angle. The method was applied to two phantoms in which the acetabular cup was implanted to the artificial bone. We acquired multiple set of MDCT image for the same phantom with changing the pelvic pose in the MDCT scanner to evaluate the reproducibility. The standard deviations of measured angles in multiple acquisitions were less than 2.0 deg for both of the inclination and the anteversion angles.

The human brain atrophies with aging. By investigating the morphological change of brain structure and comparing with the deformation of normal aging, we can diagnose the cerebral diseases such as Alzheimer's diseases, Perkinson disease, etc. Magnetic resonance imaging (MRI) is a good diagnostic tool because it is non-invasive to the human body, and it can take the thin-sectional images with high contrast. This paper shows a method to investigate the morphological structure of the human brain using MRI. The method segments the brain region, liner and non-liner registration to a standard brain, and inverse wrapping from the standard brain. By giving some landmarks to the standard brain, we can obtain the landmark position of each subject. We investigate the morphological change using the position changes. The method has been applied to open database (IXI dataset) that contains 619 subjects whose age is 19.98-83.62 years old.

In this paper, a two-staged method of determining the order of medical examinations is presented to reduce sojourn times for patients visiting a department of cardiology. The first stage works as the batch process in the nighttime, and determines the order of medical examinations for patients with appointments. The second stage is designed for patients without appointments, and searches free time slots into which the examinations for such patients can be inserted in the list of examinations that the first stage generated during the night. When determining orders, the proposed method uses values randomly generated from surveillance results as examination times. Experimental results show that the proposed method achieves short sojourn times both for patients without appointments and for those with appointments, compared with the determination manually made by a nurse.

In this paper, an approach to estimating waiting time is presented for outpatients visiting the department of ophthalmology. It determines the number of virtual outpatients according to the probability density functions of the gamma distribution. Assignments of arrival time intervals for such outpatients depend on random numbers generated from the exponential distribution. It assumes that outpatients undergo all examinations within the waiting time equal to the difference between the start time point of consultation and arrival time. The examinations are virtually managed according to six rules established by the surveys at the hospital. A proposed system requires a day of the week and arrival time of a target outpatient desiring, disease name, and name of ophthalmologist at the time of making an appointment to estimate the waiting time. It is revealed that the difference between estimated waiting time and actual waiting time is acceptable to outpatients. © 2013 Copyright TSI® Press.

Properties and performances of associative memories, based on Complex-valued Synergetic Computer (CVSC), are explored in this paper. In the proposed CVSC, the state vectors, prototype vectors and order parameters are encoded by complex values. This CVSC is extended from the conventional Synergetic Computer (RVSC) in which the order parameters produce real values in processing real-valued input and stored data. Performances of associative memories in CVSC are investigated through solving a problem of image retrievals where the input images are partially occluded, noise-Affected, or two-patterns superimposed. From the experimental results concerning the retrieval performances related to various sizes of images and different levels of defectiveness of input images, we conclude that CVSC outperforms RVSC.

This paper explores complex-valued multilayer perceptrons (MLPs) with the mechanism of stochastic resonance (SR). SR is a phenomenon such that a weak periodic signal in the system can be enhanced and detected in the presence of noises. It is expected that the combination of complex-valued encoding and SR mechanism will improve the performance of MLPs, rather than MLPs with either of them. The performances are evaluated through approximations for one-and two-dimensional functions. It is shown that complex-valued MLP with SR could achieve more precise approximations rather than conventional real-valued MLP and complex-valued MLP without SR mechanism.

The extended boundary-node method (X-BNM) has been modified for improving the accuracy degradation due to the boundary shape and its performance has been numerically investigated by comparing with the standard one. For the case where the boundary shape is strongly concave, the results of computations show that the accuracy of the modified X-BNM is always higher than that of the standard one. In addition, the speed of the modified X-BNM is almost equal to that of the standard one. Therefore, it is found that the performance of the modified X-BNM is much superior to that of the standard one. © 1965-2012 IEEE.

In this paper, a method of determining examination groups is presented for new outpatients visiting the department of ophthalmology, using support vector machines (SVM's). Assuming that interview sheets are divided into four classes, the proposed method copes with the examination determination as the classification of the sheets. The data are generated from handwriting sentences in the sheets, and they are arranged in the form of a matrix. Some nouns and adjectives in the sentences are chosen as elements of the matrix, and are assigned to columns of the matrix. The sentences in the sheets are assigned to rows of the matrix. Frequencies of the chosen words appearing in the sentences are basically given as element values in the matrix, and weighting is applied to element values associated with the words having comparatively high frequencies for sentences belonging to two classes at most. Normal SVM learning constructs a discrimination model, and defines four discriminant functions associated with the model. Since one-versus- all approach is employed, the class of data to be examined is determined according to output values of the four functions. It is established that the determination made by the proposed method achieves as favorable accuracy as the first determination made by an average ophthalmologist.

In this paper, selection of formulas to calculate intraocular lens (IOL) power is presented for a cataract patient, using support vector machines (SVM's) and genetic algorithm (GA). Each of the used data belongs to one of the three classes specified by names of the power formulas. The proposed method addresses the formula selection as the issue of classifying the data associated with the patient. The formulas have variables into which values of axial length and corneal refractive power are substituted, and hence each of the data has elements associated with them. Other elements probably useful in classifying the data are determined by GA. The final discrimination model is constructed by SVM learning using the above data. It consists of three coordinate spaces having two regions corresponding to two formulas out of three. Each of the spaces provides a potential solution. The formula to be used for some patient is specified by a majority of the potential solutions. The experimental results establish that the proposed method achieves the substantial percentage of concordance for the classification.

社会性昆虫のミツバチにおいては,8の字ダンスという特異的な行動が知られている.これは餌場の方向と距離を他の個体に伝えるための一種の「言語」であると考えられている.この行動が各個体や集団におよぼす影響を解析するためには,ミツバチ集団においてダンスを行っている個体とその行動状態を抽出する必要がある.通常,この解析は撮像した動画像から手作業で行われているが,本研究では,撮像動画像を時間軸方向に展開した断面画像を用いて,ダンスを行っているミツバチ個体を自動的に検出する手法を提案する.また,この提案手法を巣板上の実際のミツバチ集団を撮像した動画像に適用した結果を報告する.

In this paper, a method of determining examination groups for new patients is presented, using self-organizing maps. Assuming that interview sheets are divided into four classes, the method copes with the examination determination as the classification of the sheets. The data are generated from handwriting sentences in the sheets. Some nouns in them are picked up as elements of the data. After map learning is complete, its neurons are labeled. The class of the sheet corresponding to the data to be checked is specified by the label of the winner neuron for the data. It is established that the multiple-map-based scheme achieves favorable classification accuracy.

Waggle dance by honeybee workers is a behavior for sharing information concerning feeding sources. Thus it is important to analyze the occurrence timing and distribution of waggle dance. This paper presents a scheme for automatically detection of workers undergoing waggle dance from the image sequence. The performance of the proposed scheme is investigated by using actual images obtained by a video camera.

In this paper, a method of determining examinations is presented for outpatients visiting the department of ophthalmology. It assumes that each of the interview sheets belongs to one of the four classes, and copes with the examination determination as the classification of the sheets using self-organizing maps. Training data presented to the maps are generated from handwriting sentences in the sheets. Some nouns, adjectives and adverbs that ophthalmologists consider to be of comparative importance are chosen as elements of the training data. The element values basically depend on frequencies of the chosen words appearing in the sentences. After map learning is complete, neurons in the map are labeled. The data class associated with the sheet to be checked is given as the label of the winner neuron for the presented data. It is established that the proposed method achieves as favorable classification accuracy as initial determination made by ophthalmologists.

The performances of associative memory, based on complex-valued Hopfield neural network, are presented in this paper. The structures and parameters of complex networks are considered in the presented network. Retrieval performance with respect to the cluster coefficient, one of the parameters in complex networks, are explored.

Stability of embedded patterns on associative memory is investigated in this paper. The associative memory is composed of complex-valued Hopfield neural network, in which the state of neurons are encoded by the phase values on a unit circle of complex plane. Local iterative learning scheme and Projection rule are used for embedding the patterns onto the network. The retaining performance for embedded patterns are evaluated through storing randomly generated patterns and gray-scaled images with changing the resolution of neuron state.

In this paper, waiting-time estimation is presented for outpatients visiting department of ophthalmology. It determines the number of virtual outpatients according to the probability density functions of the gamma distribution. Assignments of arrival time intervals for such outpatients depend on random numbers generated from the exponential distribution. It assumes that outpatients undergo all examinations within the waiting time. The examination order is determined by six rules established by the surveys at the hospital. Once a target outpatient specifies a day of the week and arrival time of the outpatient desiring, disease name, and name of ophthalmologist at the time of making an appointment, the proposed system calculates the waiting time for the outpatient. Experimental results establish that the difference between estimated waiting time and actual waiting time is acceptable to outpatients.

A localization scheme for wireless sensor networks is presented in this paper. The presented scheme is an anchor-free scheme, in which none of geometrical information is necessary for sensor nodes. Thus, only local interaction among sensor nodes is required for estimating the locations of the sensor nodes. This scheme employs the link quality indicator (LQI) and hop count between the sensor nodes for estimation. Our experiments involve the implementation of the proposed scheme on Zigbee sensor modules. The results of the experiments indicate that estimation can be successfully carried out for networks comprising four sensor nodes.

In this paper, self-organizing-map-based video object segmentation is proposed, assuming that either Y-quantification or HSV-quantification can be systematically selected. Given a video sequence, the value of probability density function is calculated for each component value according to kernel estimation at the first fame. Some areas randomly chosen from the background are then examined, using each component value, whether it is misjudged that they include the target object. The quantification is determined so that occurrence frequency of the above false extraction can be reduced. The data presented to maps are generated, based on the selected quantification. Experimental results show that the proposed method well recognizes the target object.

In this paper, self-organizing-map-based video object segmentation is proposed, assuming that either Y-quantification or HSV-quantification can be systematically selected. Given a video sequence, the value of probability density function is calculated for each component value according to kernel estimation at the first fame. Some areas randomly chosen from the background are then examined, using each component value, whether it is misjudged that they include the target object. The quantification is determined so that occurrence frequency of the above false extraction can be reduced. The data presented to maps are generated, based on the selected quantification. Experimental results show that the proposed method well recognizes the target object.

In this paper, methods of selecting formulas to calculate intraocular lens (IOL) power for a cataract patient are presented, using support vector machines (SVM's) and self-organizing maps (SOM's). Each of training data has measured values associated with axial length and corneal refractive power as element values. Three IOL power formulas have variables into which these values are substituted. The discrimination model constructed by SVM learning consists of three coordinate spaces having two regions corresponding to two formulas out of three. Each of the spaces provides a potential solution. The formula to be used for some patient is specified by a majority of the potential solutions. The SOM-based scheme determines the formula suitable to some patient, observing the label attached to the winner neuron for the presented data having the above element values associated with the patient. The experimental results finally establish that the proposed SVM-based scheme especially works well to select the formula.

For the realization of nanocomputers it will be important to have built-in defect-tolerance, which is the ability to overcome the unreliability caused by defective components. This paper explores defect-tolerance for nanocomputers based on Self-Reproducing Loops, in which each of the loops in the system acts as a computational element, supporting the propagation of signals along transmission wires and their processing in logic elements. The loop-based design facilitates the adaptation to defects through the expansion of wires such as to prevent them from being blocked by defects. The proposed system is implemented on an asynchronously timed Cellular Automaton. © 2010 ISAROB.

A voltage-scheduling heuristic is presented for a real-time multi-processor system to reduce its energy expenditure. It consists of offline and online components. Provided that processors can do useful computation with two voltage levels, in the offline component, two static voltage-scheduling algorithms independently assign a task two time instants when the task absolutely completes its execution. In the online component, one of the time instants is chosen according to the averaged ratio of actual execution units compared to worst-case scenarios of the completed tasks, and the portion to be run at a lower voltage is extended for a task succeeding to them. Simulations are made to show the effectiveness of the proposed heuristic.

Synergetic Computers(SCs) are a type of selforganizing ones and have widely been used for pattern recognition or classification problems. In conventional SCs, the order parameters are usually encoded by real values, which determines internal states. In encoding the information of images, however, the complex or hyper-complex values are used to achieve the outperformance. In this paper, we propose Complex-Valued Synergetic Computers (CVSCs) where the order parameters are encoded by complex values. We further investigate the performances of our proposed CVSCs through the recognition of gray-scale images, by comparing it with conventional Real- Valued Synergetic Computers (RVSCs).

An implementation of Self-Reproducing Loops (SRL) on Brownian Cellular Automata (BCA) is proposed in this paper. BCA are asynchronous cellular automata in which certain local configurations propagate randomly in the cellular space, resembling Brownian motion. In the proposed SRL, the signals in the loops and the loop heads can move backward and forward because of the Brownian nature of BCA, thus making it possible to avoid collisions of loop heads. © 2010 ISAROB.

A localization scheme for wireless sensor networks is proposed in this paper. The proposed scheme is a range-free and anchor-free scheme; hence, only local interaction among sensor nodes is necessary for estimating the locations of the sensor nodes. This scheme employs the link quality indicator (LQI) and hop count between the sensor nodes for estimation. Our experiments involve the implementation of the proposed scheme on a ZigBee sensor module. The results of the experiments indicate that estimation can be successfully carried out for networks comprising three or four sensor nodes. © 2010 SICE.

Double contrast (DC) X-ray images are useful and cost-effective for the diagnosis of stomach tumors. However, it is difficult to automatically extract tumors from these images. This is due to the variations in the tumors appearing in the images as a result of the changes in stomach shapes and the distribution of barium meal in the stomach. In this paper, we propose an automated method for detecting tumors in DC Xray images. Our method utilizes the distributions of contrast gradients in the images. The performance of our method is demonstrated by using actual DC X-ray images.

In this paper, a method of selecting one type out of three types of intraocular lens (IOL) power formulas for data of each individual cataract patient is presented, using self-organizing maps (SOM's). The proposed method employs three-dimensional vectors each of which has measured values of axial length, corneal refractive power and cylinder as training data for maps. The first two values are substituted into the power formulas, while the last value is associated with the astigmatism. This paper also proposes neuron labeling that depends on postoperative refractive errors occurring under the assumption that each of the three power formulas is applied. The proposed method determines the formula to be applied to some patient, observing the label attached to the winner neuron for the presented data with the above three element values associated with the patient. The experimental results finally establish that the proposed method adequately works to select the formula.

Computers with device feature sizes of a few nanometers-so-called nanocomputers-are expected within a few decades, but this expectation is accompanied by the realization that the boundary conditions of such systems differ substantially from those of current VLSI-based computers. Prominent among the concerns is the increased degree of permanent defects that will affect nanocomputers, such as defects caused by imperfections at the manufacturing stage, but also defects occurring later, possibly even during the use of these systems. New techniques to deal with defects are called for, but given the huge number of devices involved, such techniques may need to be self-contained: they need be applicable at local levels without outside control, even while computations continue to take place. This paper proposes an important element in such techniques, i.e. the localization of defects among a huge number of devices. It employs a cellular automaton-based architecture, and uses statistical techniques combined with randomly moving configurations in the cellular space to estimate defect locations.