This article describes a watchdog program that discovers “meaningful” repair cases from a field service database. “Meaningful” cases are those judged worth probing further to prevent an epidemic of quality problems. Our system has employed the Apriori algorithm, a data mining technique which efficiently performs the basket analysis. Our system proves that this data mining technique is not only useful in knowledge discovery but is also capable of performing the database watchdog task. The Apriori algorithm automatically generates frequent itemsets from a large set of records. A frequent itemsett is an arbitrary combination of values that appear more often than a threshold “minimum support.” The algorithm often generates too many itemsets for quality engineers to review carefully in their daily work. Many itemsets don't provide sufficient information to investigate further. Hence, in order not to generate these valueless itemsets. the Apriori algorithm is modified in two ways. One way is “Basket analysis on objective and explanatory attributes”, and the other is “Itemset reduction” The advantage of our method is demonstrated with some experimental results.

抄録全体を表示

This paper discusses the method to add facial expressions to an expressionless line drawing facial caricature reflecting personal impressions. The method consists of the facial caricature drawing part and the facial expressions generating part. In the facial caricature drawing part, impressions and features of a face are inputted by the use of linguistic terms, and an expressionless facial caricature is drawn. In the facial expressions generating part, four facial expressons, i.e., laughing, sad, angry, surprising, are considered. Questionnaires about the comparison of expressive facial caricatures and expressionless ones are performed in order to obtain the parameter value change data for each facial expression. The parameter values transformation from an expressionless face to an expressive one is defined using these data for the addition of facial expressions. Expressive facial caricature drawings are evaluated subjectively and objectively. The validity of the process is discussed.

抄録全体を表示

Amount of data for operators is increasing, but the amount of information, which operators obtain is not increasing because conventional human interfaces only focus on human foreground awareness. Users usually obtain information with both of background and foreground awareness. This paper proposes a new human interface style (awareness oriented human interface), which exploits the human background awareness. The human interface promotes the operator's abilities of recognition by, 1) providing information for both of foreground and background awareness simultaneously, 2) information navigation between foreground and background awareness, 3) awareness oriented information processing, 4) supporting public awareness for multiple operators.

抄録全体を表示

This paper presents a method for integrating gamma correction and gain control functions on a CMOS image sensor with logarithmic response. The proposed method is based on a fact that gain variation and level shifting in logarithmic compressed domain are equivalent to the gamma correction and the gain control, respectively, in the exponentially expanded domain. The prototype chip integrating proposed function is fabricated with triple-metal double poly-silicon n-well 0.6μm CMOS technology. As a result of calculation using measurement data of test circuits, the gamma value can be adjusted to 0.45 when the gain given in the logarithmic compressed domain is 0.741, and a gain of 20dB is obtained by the voltage level shift of 60mV.

抄録全体を表示

This paper deals with positional information sharing in a multiple autonomous robot system, in which a numerous robot agents wander in a wide field to search for and to gather objects. To accomplish such a task effectively, it is essential for all the agents to share common information including locations of objects, comrade agents, and the place to deposit the objects. The authors have proposed a method called “COM-PASS” to realize this only using two functions of each agent - dead-reckoning and one to one bidirectional proximity communication. In this paper, authors introduce a new concept “Figure Of Confidence (FOC)” to the COMPASS. FOC is a self-estimated measure for accuracy of information in each agent. Using FOC can accelerate and improve accuracy of information sharing in COMPASS. Moreover, a new concept named “Object Horizon (OH)” is derived from FOC in this paper which indicates a possible area of existence of an object. OH provides an effective means to handle multiple objects without complexity. The validity of FOC and OH are confirmed through computer simulations and experiments.

抄録全体を表示

A new CGH (Computer Generated Hologram) compressed and transmitted and reconstructed system (CCTRS) has been established in which JPEG baseline-encoding model and Fresnel transforming technique (FTT) have been adapted. This system can be applied in the remote signal processing using the digital filter of CGH. When the compression ratio is achieved to 1.5303%, the information of processed CGH can be effectively reconstructed using FTT. Moreover, in this system the size of image displayed is bigger (i.e., 18.07×18.07(cm²)), the system structure is simpler, the system costs lower than that using the electron holographic hardware display system, and can be widely applied by computer operators. In experiments, in terms of the holographic principle of E. N. Leith and J. Upatnieks, a CGH has been made with a computer, the information of processed CGH has been analyzed and the information distribution of processed CGH has been compared with the one of the original CGH. The reconstructed image quality of processed CGH has been discussed. Finally, Compression ratio (R), Mean squared error (MSE) and Peak signal to noise ratio (PSNR) have been precisely calculated and analyzed to evaluate the image quality of processed CGH and the reconstructed image quality of processed CGH. And the noise influence of the system for processed CGH has been analyzed and discussed in detail. In CGH compressed and transmitted, the reason has been carefully explained why the error problems can be caused as the pixel's amplitude and phase information of CGH varied. This method of processing CGH has been effectively verified by experiments.

抄録全体を表示

Computer hardware testing is performed with test, programs generated by using machine instructions or high-level language. A parallel computer is characterized by the complexity resulting from its system configuration consisting of an array of a number of processors, and also by the parallel processing unit whose hardware configuration can be varied corresponding to the system objects and the required performance conditions. Because of this, it is required to prepare the test. programs corresponding to the respective hardware configurations. This means that the number of test programs required increases with the number of hardware configuration types designed, and this requires a tremendous amount of labor for their generation.
This paper proposes a system for efficiently generating the test programs to be used for hardware testing of the scalable SIMD parallel computer. This system makes the most of the functions and features of the scalable SIMD parallel computer, and generates the test programs without depending on the hardware configuration of the parallel processing unit. By using this system, it is possible to reduce the types and number of the test programs, and consequently, the period for the development.

抄録全体を表示

Simulated Annealing (SA) is known as one of useful heuristic optimization techniques. However the performance of standard SA depends on an initial state at starting temperature because of one-point search of SA. In this paper, to overcome the dependency on the initial state, we propose Immunity-based SA (ISA) that has some features of immune system, combining clonal selection, immunology memory, and idiotypic network with Multipath SA (MSA) which searches the solution space for optimal solutions in parallel starting from many initial states. ISA is expected to improve the local and global search ability and maintain the diversity of the population. We demonstrate the efficiency of ISA by applying it to the Quadratic Assignment Problems. Experimental results show that ISA performs better than SA and MSA.

抄録全体を表示

Test feature classifiers are generated directly from training samples and have a 100% recognition rate on training data. Although this perfect learnability is an important feature of the classifiers, it does not guarantee a good generalization. In this paper, we concentrate on the performance of classifiers on test data, and describe cases when a 100% recognition rate can be achieved. We show that training data can contain information about possible discriminant boundaries between entire classes. In general, it is impossible to extract this information, although we propose a heuristic algorithm which could lead to a 100% recognition rate. To test the performance of the classifiers, we apply them to both artificial and real data. For the real data, we use the well-known breast cancer and satellite image databases. Our experimental results show that the proposed classifiers have not only a high recognition ability, but also confirm the ability of a 100% recognition rate in real classification problems.

抄録全体を表示

In order to segment robustly character areas of serial numbers in weak-contrast and noisy license plate images which are extracted from the front view of cars by the method described in our previous paper, we propose a useful method which is constructed by artificial neural networks corresponding to each digit of the serial numbers. The ability of proposed method is compared to that of a histogram-based method which is used as an example of conventional methods. In the histogram-based method, touching characters and noisy characters are often mis-segmented. The other hand, the network-based method can segment precisely even for those characters. For 595 license plate images, a segmentation rate of 96.7% in the histogram-based method and a rate of 100% in the network-based method with an additional second search are gained. Then, a network-based recognition method for the segmented characters is investigated. By randomly varying the segmentation size of trained characters in learning process of the network, all the serial numbers in the 595 images can be recognized.

抄録全体を表示

In this paper, PN (Petri Net) is used for the qualitative modeling of SFC (Sequential Function Chart) rules of evolution with the cyclic scan of the programmable controller. With the proposed PN representation method, it is clarified that the SFC rules of evolution, such as the relationship between the active step and the applicable action, can be easily and systematically examined and analyzed, although the proposed PN is much complicated as compared with the general PN representation. Moreover, the relationship between the alternative and simultaneous sequences and the execution time of SFC is examined by the PN behavioral analysis. The experimental verification for the theoretical prediction shows that the execution time is independent of the number of the alternative sequences, but dependent on that, of the simultaneous ones.

抄録全体を表示

We experimentally investigated the influence of laser beam reflected at the surface of substrate on single beam fiber trapping. From these results, we verified that three-dimensional optical trapping of a micro-sphere dispersed in ethanol solution could be obtained by the laser beam reflected at the surface of substrate.

抄録全体を表示