A basic study on the Person Identification system based on writings

1)我们设计了一种新的程序，称为“局部弧模式方法”，作为个人识别的辅助手段。它是在一个系统中实现的，在这个系统中，一个人的身份是通过他的作品来确定的。任何出现在这个系统前的人都必须以这样的方式写一个指定的日语句子，句子中的任何字母都是一个接一个单独写的。然后将每个书写字符转换为二进制模式，并测量了某些特定模式的频率，这些模式是我们过去提出的弧形模式的简化形式。这个人是根据这些频率与Mahalanobis的距离来识别的。在这个系统中，实现了本地化的思想，使系统小，在这个意义上说，所需的内存和时间远远小于过去提出的其他系统，2）为了将上述系统实现为个人计算机系统，我们编写了一个软件，该软件可以使用图像扫描仪将书面文档读入个人计算机。由于个人在书写中的个人倾向性在这个系统中起着至关重要的作用，我们试图通过重建能够很好地代表个人倾向性的文字来形象化这种个性。4）在实现的基于书写的小规模个人识别系统中，出现了一个包含以下四个部分的菜单：1.字符数据输入部分，2.特征提取部分，3.生成字典部分，4.识别部分。但是由于微机规模的限制，MS-DOS上的MS-FORTRAN语言作为一个系统，内存容量有限，使得系统的实现并不理想。5）本研究已经或准备向IEEE的PAMI或Pattern Recognition等相关期刊投稿。我们在IECIE会议或第九届ICPR上读到了一些论文。

1) We devised a new procedure called the 'Localized Arc Pattern Method' as an aid for person identification. It is implemented in a system where a person is identified through his writings. Any person who appears in front of this system is required to write a specified Japanese sentence in such a manner that any letter in the sentence is written one by one separately. Then each written character is transformed in a binary pattern and measured the frequencies of certain specified model patterns which are simplified forms of the arc patterns proposed by us in past. The person is identified with the Mahalanobis' distance based on these frequencies. In this system the idea of localization achieved to make the system small in the sense that the necessary memory and time is far smaller than other systems proposed in past, though it retains a high average correct identification rate.2) To realize the above mentioned system as a personal computer system we programmed a software which can read written documents into a personal computer using an image scanner.3) Since the individual personal tendency of distortion in writing plays an essential role in this system we tried to visualize this individuality through reconstruction of characters which well represent the personal tendencies of writings.4) In the realized small scale personal identification system based on writings a menu with the following four parts appears: 1. Character data input part, 2. Feature extraction part, 3. Generation dictionary part, 4. Identification part. But the limited scale of the personal computer made the system unsatisfactory for us in the implementation because the MS-FORTRAN on MS-DOS can not afford go much memory as a system.5) We submitted or are going to submit some papers in this study to related journals such as the PAMI of IEEE or the Pattern Recognition. We read some papers in IECIE conferences or the 9th ICPR.