Takashi Kitagawa

本稿では，メディアコンテンツを対象とした統計的一般化逆作用素の構成方式を示す．これは，与えられた印象を表す単語とその重みで表される印象メタデータからメディアコンテンツを生成する機構を実現するものである．本方式は，従来，我々が提案してきた手法であるメディアコンテンツから印象を表す言葉をメタデータとして抽出する，メディアコンテンツを対象としたメタデータ自動抽出方式手法の逆演算として構成される．しかしながら，本逆演算においては一般的に，不良設定問題が発生する．この不良設定問題を解決するために，制約条件としてそのメディアの種類に応じた統計情報やそのメディアを対象とした研究成果・理論を用いる．これにより，メディアコンテンツと言葉の間の相互変換を可能とする．さらに，本方式を楽曲メディアコンテンツに適用し，印象語で表現された印象メタデータから楽曲メディアコンテンツを生成するシステムを実装する．これを用い

It is very important that selecting what kind of features in the field of image recognition. We provide a method which extract several features of a figuration from an image. We present the metadata bringing together these features. This paper shows detailed computing method for each feature. And, we conduct an assessment experiment to show their availability. Finally we conduct an experiment in clustering using proposal metadata. As a result, we can get clusters which are similar each other about a figuration.

When one write a program of automatic composition system, music features are generally used, which were extracted from some existing music. Music features are different on each system. In this paper, to consolidate features, we define a lot of music features. Then, we make a model which select dynamically music features of the musical attribute. We experiment with this model for features-extraction of an instrument and a genre. We got ranked features of how much music attribute's identity the feature express. By using higher ranked music features, we can compose a music which has features of the musical attribute.

This paper presents an implementation of drawing support application utilizing database for defining the relation between color and impression. We have realized Mathematical model of meaning that be able to compute semantic correlations between words and Media-lexicon Transformation Operator that be able to extract weighted words from media data using knowledge database for various media. Using both methods, we have realized a connecting method among heterogeneous media data. In this paper, we implement a drawing support application that suggest impression words computed by using method previously described with "Color Image Scale" (the database written about the relation between color and impression ). This application operates with browsing software. Users draw in monochrome line art. First,users input a combination of words as impression which users want to express, then suggested some colors corresoonding to this combination. And while users draw art,this application suggests a list of impression words which understand current art. This application implemente layman leverage specialist knowledge about color without regard for it.

In this paper, we present a phoneme-impression transformation operator construction method for an arbitrary word. This method extracts impression that comes from sounds of arbitrary words, including undefined words, unknown words, and new words. It realizes interfaces between persons and computers utilizing user's feeling expression by undefined/unknown words. This paper presents a Kansei retrieval system by utilizing this method for music MIDI data as an experiment system and an application example. Finally we perform various experiments on the system in order to verify the effectiveness of this method.

This paper presents an automatic music composition method corresponding to the impression of arbitrary Japanese words by using the impressions received from the sound expression of the words. Founded on the "onso" theory, we have realized an automatic extraction method which extracts impression words as metadata corresponding to the impression by the sound expression of arbitrary words. And we have also realized a method of an automatic composition method of music, by searching for correlations between 8 groups of words about musical impressions and 6 musical structural elements by using the reserches done by Hevner. In this paper, we realize the method of automatic composition of music corresponding to the impression of the sounds of arbitrary words by connencting the impression words extraction method and the method of automatic composition of music on meta-level by using a mathmatical model of meanings.

In this paper, we present an automatic composition method of sound expression of a word corresponding to the impressions of arbitrary words. We have proposed a method of automatic metadata extraction corresponding to the impression of the souds of arbitrary words by the research called "Onso", and we propose an inverse operation of the automatic metadata extraction in this paper. This method enables a connection of any words and mediadata represented in impression words.

This paper presents an automatic music composition method corresponding to the impression of arbitrary Japanese words by using the impressions received from the sound expression of the words. Founded on the "onso" theory, we have realized an automatic extraction method which extracts impression words as metadata corresponding to the impression by the sound expression of arbitrary words. And we have also realized a method of an automatic composition method of music, by searching for correlations between 8 groups of words about musical impressions and 6 musical structural elements by using the reserches done by Hevner. In this paper, we realize the method of automatic composition of music corresponding to the impression of the sounds of arbitrary words by connencting the impression words extraction method and the method of automatic composition of music on meta-level by using a mathmatical model of meanings.

In this paper, we present an automatic composition method of sound expression of a word corresponding to the impressions of arbitrary words. We have proposed a method of automatic metadata extraction corresponding to the impression of the souds of arbitrary words by the research called "Onso", and we propose an inverse operation of the automatic metadata extraction in this paper. This method enables a connection of any words and mediadata represented in impression words.

In this paper, we present a method of automatic metadata extraction corresponding to the impression by the sound expression of arbitrary words. This method can extract impression words corresponding to the "Onso" from words or phrases. By this method, an arbitrary word given by a user can be transformed to a vector by impression words corresponding to "Onso". This method realizes the search function for media data corresponding to user's impression by using arbitrary words. In addition, this method can realize the connections between arbitrary words and heterogeneous media data.

In this paper, we represent an automatic generation method for music data corresponding to impression of color information. This method can generate a music data by measuring relations between impression words extracted from color information and 8 categories of impression words given by a psychological research. This method can realize connections between heterogeneous media such as image and music. This method can realize communication places for compound media. When this method can realize, computers may be able to share and amplify Kansei with human each other.

In this paper, we present a method of automatic metadata extraction corresponding to the impression by the sound expression of arbitrary words. This method can extract impression words corresponding to the "Onso" from words or phrases. By this method, an arbitrary word given by a user can be transformed to a vector by impression words corresponding to "Onso". This method realizes the search function for media data corresponding to user's impression by using arbitrary words. In addition, this method can realize the connections between arbitrary words and heterogeneous media data.

In this paper, we represent an automatic generation method for music data corresponding to impression of color information. This method can generate a music data by measuring relations between impression words extracted from color information and 8 categories of impression words given by a psychological research. This method can realize connections between heterogeneous media such as image and music. This method can realize communication places for compound media. When this method can realize, computers may be able to share and amplify Kansei with human each other.

CG(Computer Graphics)やCAD(Computer Aided Design)では,三次元計測装置から得た離散点群から3Dオブジェクト曲面を再構成することがしばしば要求される.この表現法として,近年,陰関数曲面が注目されている.また,離散点群上の色情報を同時に取得できる計測装置も実用化されており,再構成した陰関数曲面上に色情報を付加することが問題となっている.本研究では,同問題の新かな表現法として,MPU(Multi-level Partition of Unity)法に基づく色関数の生成手法を提案する.本提案法では,小領域で定義される区分的な色関数を構成し,重み関数を介してそれらを滑らかに結合することで全領域で定義される色関数を構成する.また,小領域セル内の色情報の状態により色関数の構成法を適応的に変えることで色境界の表現も実現される.

In order to realize associative search for a specific field by the mathematical model of meaning, it's necessary to construct a retrieval space. A construction method of retrieval space has been proposed that creates a data matrix by using a dictionary or a term dictionary. However, it's difficult to construct a retrieval space without them. This paper presents a new construction method of a retrieval space based on the relation between each word in documents. This method constructs a retrieval space by the relevance between each word calculated from word distributions in documents. By this method, a construction of a retrieval space for an associative search is performed automatically without any dictionary.

In this paper, we present an integration method of retrival spaces. This method make it possible to integrate retrival spaces based on the relation between words. In order to realize search method utilizing a vector space model for a specific field, it is necessary to construct a retrival space for the field. The number of the vocabularies, which the retrival space constructed from a index of a document can express is restricted. The problem is solved by this method that uses other retrival spaces. It is thought that the retrival space constructed method by the index part of the documents containing this method is applicable to mediadata and document search for broad fields, such as a field of not only a scientific field but a hobby.

In order to realize associative search for a specific field by the mathematical model of meaning, it's necessary to construct a retrieval space. A construction method of retrieval space has been proposed that creates a data matrix by using a dictionary or a term dictionary. However, it's difficult to construct a retrieval space without them. This paper presents a new construction method of a retrieval space based on the relation between each word in documents. This method constructs a retrieval space by the relevance between each word calculated from word distributions in documents. By this method, a construction of a retrieval space for an associative search is performed automatically without any dictionary.

In this paper, we present an integration method of retrival spaces. This method make it possible to integrate retrival spaces based on the relation between words. In order to realize search method utilizing a vector space model for a specific field, it is necessary to construct a retrival space for the field. The number of the vocabularies, which the retrival space constructed from a index of a document can express is restricted. The problem is solved by this method that uses other retrival spaces. It is thought that the retrival space constructed method by the index part of the documents containing this method is applicable to mediadata and document search for broad fields, such as a field of not only a scientific field but a hobby.

本論文では, 特定分野を対象とした書籍の索引を利用することで単語間の関連によるメタデータ空間を生成する方法を提案する.特定分野を対象とした意味の数学モデルを用いた検索機構を実現するためには, その分野を対象としたメタデータ空間と呼ばれる検索空間を生成する必要がある.これまで, メタデータ空間は, 用語辞典, 専門的な知識を用いて生成していた.また, 対象とする分野に用語辞典が存在しない場合, メタデータ空間を生成することが難しい.本方式では, 用語辞典を必要とせず, 専門的な知識なしでメタデータ空間を生成可能とする.更に, 意味の数学モデルに本メタデータ空間生成方式を適用することにより, その特定分野のドキュメントやメディアデータなどの検索対象データを対象とした単語間の関連に基づく検索機構が実現できる.本論文では, IT分野を対象とし, 用語辞典によるメタデータ空間と索引によるメタデータ空間で画像メディアデータ検索を適用し, 比較を行う

三次元領域において, 重複のない離散点の集合から三次元形状(3Dオブジェクト)を再構成する問題を考える.本研究ではこの問題において, オブジェクトを陰関数によって表現する方法を用いる.同方法では, 連立一次方程式を解くことで関数を決定するが, CSRBF(Compactly-Supported Radial Basis Functions)と呼ばれる基底を用いた場合, 係数行列は疎行列になる.本研究では, この連立一次方程式をCSRBF型連立一次方程式と呼ぶことにし, 同連立一次方程式の解法について考える.従来, 同連立一次方程式の解法には係数行列の構造に特化した直接法が用いられてきた.しかしながら直接法による解法は, 連立一次方程式の大規模化に伴い演算コストの増大を招く可能性がある.そこで本研究では, CSRBF型連立一次方程式の解法に前処理付反復法を導入する.本研究の目的は, CSRBF型連立一次方程式の解法として前処理付反復法を適用し, 求解に要するCPU時間の高速化を図ることにある.数値実験により, 同連立一次方程式の

In order to realize associative search for a specific field by the mathematics model of a meaning, it is necessary to establish the retrieval space called metadata space for the specific field. A metadata space was established using a dictionary, a term dictionary, and special knowledge. However, when neither a dictionary nor a term dictionary existed in the target field, it was difficult to establish metadata space. This paper presents a new construction method of a metadata space based on the locality of words in documents. This method enables establishment of a metadata space which measures the relation between words easily without making any dictionary. The words related associative search for documents and mediadata of a specific field is realized by applying the metadata space to the mathematics model of a meaning. This paper shows the experimental results which applied the mathematics model of a meaning to the metadata space established using this method.

This paper presents the system which extracts metadata automatically for voice media data. Specifically, the mounting system of Media-lexicon Transformation Operator ML is shown. First, using research of Banse and Scherer, a sound is analyzed and 29 voice parameters are obtained. Next, correlation with a voice parameter and an impression word is asked for an impression word with weight. The obtained weight is metadata and the impression which man receives from a sound is shown.

This paper presents an implementation method of associative heterogenous mediadata search for music data and image data. This method is realized by measuring correlation of the impression words extracted from media-lexicon transformation operator (automatic metadata extraction method) for each mediadata. The feature of this method is to bridge over heterogeneous mediadata which exist independently as different database resources.

In order to realize associative search for a specific field by the mathematics model of a meaning, it is necessary to establish the retrieval space called metadata space for the specific field. A metadata space was established using a dictionary, a term dictionary, and special knowledge. However, when neither a dictionary nor a term dictionary existed in the target field, it was difficult to establish metadata space. This paper presents a new construction method of a metadata space based on the locality of words in documents. This method enables establishment of a metadata space which measures the relation between words easily without making any dictionary. The words related associative search for documents and mediadata of a specific field is realized by applying the metadata space to the mathematics model of a meaning. This paper shows the experimental results which applied the mathematics model of a meaning to the metadata space established using this method.

This paper presents the system which extracts metadata automatically for voice media data. Specifically, the mounting system of Media-lexicon Transformation Operator ML is shown. First using research of Banse and Scherer, a sound is analyzed and 29 voice parameters are obtained. Next, correlation with a voice parameter and an impression word is asked for an impression word with weight. The obtained weight is metadata and the impression which man receives from a sound is shown.

This paper presents an implementation method of associative heterogenous mediadata search for music data and image data. This method is realized by measuring correlation of the impression words extracted from media-lexicon transformation operator (automatic metadata extraction method) for each mediadata. The feature of this method is to bridge over heterogeneous mediadata which exist independently as different database resources.