Development of the computer environment for the usage of online handwritten symbols

为了开发一个能够有效利用在线手写符号的系统，我研究了以下两个课题。首先，提出了一种利用由特定用户书写的在线字符合成的虚拟样例来改进从样例中学习到的人物特征分类器的方法。为了获得好的分类器，通常需要一个包含足够数量的手写字符变体的大型数据库。在这项研究中，我提出了一种基于特定用户书写的在线字符的人工扩充样本来训练支持向量机进行字符识别的方法。在线字符通过对每个笔画进行SAE变换来合成虚拟示例。利用人工生成的模式训练支持向量机分类器。通过手写日文平假名字符分类实验验证了该方法的有效性。第二，为了便于新手作曲家创作一首音乐，提出了一种利用现有的音乐数据库扩展用户输入的音乐的方法。该过程的概要如下：1.在数据库中，通过基于其基音序列比较URR输入的音乐和数据库的所有部分，找到最相似的部分；2使用线性回归模型，计算检索的音乐和用户输入的音乐之间的变换矩阵，3.通过对检索的音乐之前和之后的部分应用计算的矩阵的变换，可以产生新的音乐；以及4.将产生的音乐添加到用户输入的音乐之前和之后的部分。在我的实验中，我们使用了一个由12首音乐组成的数据库，创作了几首欠我们的音乐。结果，该工具被证实可以产生舒适的音乐。

In order to develop a system in which online handwritten symbols can be utilized efficiently, I have studied the following two subjects. First I propose a method to improve the personal character classifiers which am learned from examples by using the virtual examples synthesized from on-line characters written by a particular user. To obtain the good classifiers, usually a large database which contains enough number of variations of handwritten characters is required. In this study, I have proposed a method to train SVM for character recognition based on the artificially augmented examples using on-line characters written by a particular user A virtual example is synthesized from an on-line character by applying Sae transformation to each stroke. SVM classifiers am trained by using the artificially generated patterns. The effectiveness of the proposed method was examined by experiments of handwritten Japanese Hiragana character classification Second, to make it easier for a novice composer to produce a piece of music, a method which expands a user-entered music using an existing music database, has been proposed. The outline of the procedure is as follows: 1. in the database, the most similar part is found by comparison between the urer-entered music and all parts of the database based on their pitch sequences, 2 using linear regression model, a transformation matrix between the retrieved music and the user-entered music, is calculated, 3. new music can be produced by applying the transformation with the calculated matrix to the parts before and after the retrieved music, and 4. the produced music am added to the parts before and after the user-entered music. In my experiments, we composed several Owes of music with the use of a database which consists of twelve pieces of music. As a result, it was confirmed that the tool could produce comfortable music.