In this paper, we propose a method of a relationship analysis between cephalometric radiograph shape data and swallowing ability data. In order to acquire new knowledge of the relationship, the shape should be represented without missing features. In addition, a method of the relationship analysis between two data sets which are observed by different ways, an X-ray image and swallowing ability tests. Quantization and correspondence of points by SOM2 are employed for shape representation. A common map of CCA-SOM are employed for the relationship analysis method. The propose method provides relationships between cephalometric radiograph shape and swallowing ability.
The purpose of this research is to develop a self-organizing map (SOM) which comprehensively visualizes the overall picture of multi-view relational data. In order to realize this, we extended SOM to multi-view data and made it possible to estimate the factors common to all views. This algorithm is regarded as a nonlinear extension of canonical correlation analysis by SOM. Furthermore, we tried to incorporate the developed multi-view learning algorithm into the SOM for relational data. We applied our method to wine data analysis and showed its usefulness.
Attempts to detect the early stages of glaucoma by image processing techniques have been proposed. In real medical data, differences in classes may exceed the differences between classes, and the selection of feature and the preprocessing become important. In this paper, we propose a new method for fundus image analysis by using filter bank with the performance evaluation function based on the significance by SOM. With using of the significance by SOM, we will clarify the effect of filtering affecting the classification performance of fundus image analysis using filter bank.
We propose a method for supporting inspiration of the means for solving problems through visualization of technical solution described in the patent document data. First, representative words are extracted to generate word level co-occurrence probability vectors. Then, the correlation coefficients of the generated co-occurrence probability vectors are merged into correlation coefficient vectors, and inputted to SOM. We compare the SOM drawn by co-occurrence probability vector with the SOM drawn by correlation coefficient vector. It shows the potential of the method in supporting innovation acceleration through the extraction of an important related factors in the new technology development.
In the medical field, inspection systems using the face have been utilized. However, in such a system, it is possible that the combination of a patient’s preference and presentation face image will greatly affect the test results. Therefore, it is important to analyze the various factors that affect the impression of the face. Previous studies report many preference factors that are common among subjects. In this study, we analyze preference factors that are different for each subject by evaluating Japanese faces in the hedonic dimension. We focus on the face of the subject’s own face is one of these and perform a quantitative analysis using six features of the eyes, mouth, and nose. In addition, we perform a quantitative analysis using six features of the eyes, mouth, and nose. The experimental results show that it is important to focus on the similarity between a subject’s eyes in a positive face and their eyes in a negative face. Moreover, we analyze the relationship between degree of preference and facial expression recognition using the effective index of the self-face. As a result, we suggest that faces with eyes that are similar to the self-face are evaluated positively in the expression recognition.
Artificial bee colony (ABC) algorithm is a swarm intelligence algorithm which was inspired by foraging activity of honey bees and an approximate optimizing technique aiming at the real-valued optimization. ABC algorithm has high search performance to the various types of application, however, it includes some problems. For example, there is a problem that ABC algorithm is slow to converge to good solution because it makes the search process with high regard for the diversity of individuals. In recent years, researches on the advanced method of ABC algorithm have been performed briskly, and many hybrid methods which took in the idea of other evolutionary computing method are proposed. In this research, we propose Arithmetic Crossover based ABC algorithm (AC-ABC) which is the advanced method which raised search speed by including arithmetic crossover which is one of the crossover method used by real-coded GA in search processing of ABC algorithm, and Global Search type ABC algorithm (GS-ABC) which raised search performance by using the stochastic search processing used for crossover and mutation in GA into the part of variable selection process in ABC algorithm. We performed the function optimization simulation to confirm the efficiency of proposed method, and it is proved that GS-ABC shows higher search performance than the conventional method in all of six benchmark functions.