An Intelligent Character Recognition System with High Accuracy and High Speed by Integrating Image-Type and Logical-Type Information Processings

1.通过这个研究项目，开发了一个快速和准确的智能识别系统SEIUN （系统化利用智能超高速神经系统）。在系统中，使用直接元素特征向量的字符识别算法是实现的。图像--类型和逻辑类型处理系统的系统考虑因素。在级联中连接的几个模块的前称为IP系统的因素。每个模块被设计为系统阵列处理，用于规范化、构造、特征提取、距离计算和关联匹配。一些基本处理单元被设计为LSI芯片。实现IP系统的目标是Fujitsu Co. Ltd.和Fujitsu Tohoku Digital Technology Co. Ltd.，IP系统的最终版本158个字符/秒的识别速度得到了实现，该研究的满意度是一个目标速度。由18个印刷板组成的硬件组成，带有6层A3尺寸，每个尺寸约450个IC芯片AR ... More e Laid. 3。在结构分析中，什么是逻辑的内核--类型处理，一个由实验开发和验证的kanji字符线段基本结构的描述方法。关于逻辑上的修复问题--类型处理可能需要通过使用SEIUN.4来解决。两种类型的分段算法被提议用于文档的识别和分段的准确性被实验所覆盖。其中一个算法是在IP系统中实现的，它对提高IP系统的速度有影响。Furthermore，用于在IP系统中开发和实现被删除的字符和手工写的字符的新的识别算法。良好的认知表现得到了实验的证实。自从特征向量之间的距离计算被用于语音和/或图像识别，以及对IP系统的修改已经计划在设计阶段。这个系统是为了语言的识别而完成的，是为了字符识别而完成的。Less（低）

1. By this research project, a fast and accurate intelligent recognition system SEIUN (Systolic Elaborated Intelligent Ultraーhighーspeed Neosystem) was developed. In the system, character recognition algorithms using directional element feature vectors are implemented. The system consists of imageーtype and logical-type processing systems. The former called IP system consists of several modules connected in cascade. Each module is designed as a systolic array processing one of normalization, thinning, feature extraction, distance calculation and associative matching. Some of elementary processing units are designed as LSI chips.2. Implementation of IP system was entrusted to Fujitsu Co. Ltd. and Fujitsu Tohoku Digital Technology Co. Ltd. For the final version of IP system 158 characters/second of recognition speed is achieved, which satisfies a target speed of this research. The hardware consists of 18 printed board with six layers of the A3-size on each of which about 450 of IC chips ar … More e Laid.3. On the structural analysis, which is a kernel of logicalーtype processing, a description method of basic structure of line segments of kanji characters was developed and validated by experimentation. Remaining problems on the logicalーtype processing are expected to be solved by using SEIUN.4. Two kinds of segmentation algorithms were proposed for a document recognition and the accuracy of segmentation was ascertained by experimentation. One of the algorithms was implemented in IP system and it contributes to the speed up of IP system. Furthermore, new recognition algorithms for defaced characters and hand-written characters were developed and implemented in IP system. The good recognition performance was confirmed by experimentation.5. Since calculation of distance between feature vectors is used in common for speech and/or image recognition, modification of IP system had been planed in design stage. The system was completed as the one for speech recognition of words as well as character recognition. Less