In this paper, a method of determining examinations is presented for new outpatients visiting the department of ophthalmology, using support vector machines (SVM's) and self-organizing maps (SOM'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 each sheet are assigned to a row of the matrix. The element values basically depend on values associated with frequencies of the chosen words appearing in the sentences. The proposed method uses rows as training data, and constructs a discrimination model, based either on normal SVM learning or on normal SOM learning. The SVM-based method defines four discriminant functions associated with the model. Since one-versus-all approach is employed, the class of data associated with the sheet to be examined is determined according to output values of the four functions. The SOM-based method labels neurons in the model (map) after normal learning is complete. The data class 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 an average ophthalmologist working at the leading hospital.
In this study, we propose a non-contact input device using only eye movement without the calibration for correcting the gaze direction. The traditional input device using eye positional information requires head movement. However, the severely disabled people who are unable to move their head, e.g. amyotrophic lateral sclerosis, are difficult to use this device. Although an eye-gaze input device does not require the head movement, the calibration to correct gaze direction induces the degradation of usability. The proposed system allows users to control a mouse cursor of a personal computer using only eye movement. The mouse cursor control method using the distance between the center position of pupil and reference point eliminates the need for head movement and calibration. The experiments were performed to evaluate the operational performance and usability of this input device. From the experimental results, the operation time became shorter with the increase of trial numbers. In addition, the parameter setting for respective individual improved the operational performance of the input device.
As an approach for visually impaired students to learn science, we are developing a system which makes them perceive scientific things, e.g., mathematical 3D objects and dynamical phenomena, by means of a haptic device. This paper describes the first attempt towards the construction of the system. In our system, the touching model application, that is, the program which enables them to recognize the virtual reality of a scientific thing controlled by the haptic device, and blind students can experience the thing by haptic sense. They can, thereby, have mental images of scientific objects and phenomena which plays an essential role in understanding science. As the preliminary examination, we made some trial touching models and got the blind subjects to try them in order to clarify the validity and the problem of our system. We explain these touching models and the results of these experiments as well as the valuable comments from the subjects. Furthermore, the directivity off future developments of the system is addressed based on the knowledge acquired from this experiment.
Tactile graphics are images that use raised surfaces so that a visually impaired person can feel them. Tactile graphics are necessary for visually impaired students when they study mathematics and science. Since producing tactile graphics is not simple task, an intelligent computer-aided system for assisting the production of tactile graphics is needed. Mathematical graph recognition from printed materials plays an important role in developing such a system. So, this paper focuses on part of a method of mathematical graph recognition. A mathematical graph consists of character strings, mathematical formulas and graph elements such as the rays representing the x-axis and the y-axis, and the straight lines or the curves representing functions or equations. Graph elements are drawn not only by solid lines, but also broken lines. This paper discusses a method for extracting and recognizing graph elements from mathematical graphs. The effectiveness of our method is evaluated by computer experiments.
In this paper, based on the dissemination of the portable brain machine interface, the personal authentication is investigated by using EEG signals during auditory stimulations in outdoor environment. Seven healthy subjects participated in this experiment and both rare and frequent auditory stimulations were presented using oddball paradigm in indoor and outdoor environment. The event related potentials were observed in target stimulations, whereas not observed in standard stimulations. By applying machine learning technique the personal authentication rate is derived and the feasibility of outdoor EEG-based personal authentication is discussed.
In this paper, we propose an image estimation method with evaluation expression and its application to attentive hearing dialogue system. The aim of this paper is to encourage and satisfy the desire for conversation of the elderly. In order to promote the desire, there are two important points: estimation of feeling of the elderly from their utterances and creation of sentences considering their feelings. In the proposed system, sentences are created based on the estimation of impression of the words contained in the input sentences. For the impression estimation, several language resources are employed and combined to create new dictionaries: a dictionary containing about 7,000 words expressing appraisal and that containing 200,000 nouns. In the process of sentence creation, the concepts in the method of attentive hearing are reflected to encourage the desire for conversation. We carried out two kinds of experiments to evaluate the proposed system in terms of Grice's axioms and confirmed the effectiveness of the system.
In this paper, we propose a dialogue system with humor using multiple linguistic resources. The following two points are considered in this system. First, proper nouns are integrated into concept dictionaries. In particular, words in Wikipedia are added to two dictionaries, Japanese WordNet and GoiTaikei. In addition, concepts of the dictionaries are merged into declinable words by using Kyoto University's case frame. Second, humor is considered in the dialogue system. Based on the technique used in comedies, pun, comparison and kabuse, repetition of funny expressions, are employed in the dialogue. The proposed dialogue system analyzes user's utterance and extracts surface cases, predicates, and topic words. Next, the system chooses an utterance pattern and determines whether a humor is used or not from the features of input sentence and previous utterance patterns. We evaluated the proposed system in the case of humor is used and in the case not used. As a result, we confirmed that the proposed system with humor is able to make various responses and makes users more satisfied. In addition, the system is able to make a fast response due to the independence of Web.
Recently, learning environments with e-learning or with gamification are developed. Since learners have been familiar with on-line communicationthrough Social Network Services (SNS), it is expected that learning environment with network is ready to be available. In this study, a learning support environment by on-line competition quiz system is proposed. That is, learners can have high motivation to learn with ranking of experience points and human-human competition. Quiz is set as four multiple choices to continue learning easily and to be prepared various question sets through the Internet.