PURPOSE: To reduce overlapped amounts, and to improve group applying precision by applying a pattern clustering group center to a group. ;CONSTITUTION: The plural leaning patterns of one character are ununiformly divided (ST1), and the characteristics of the learning patterns are extracted (ST2). Then, the constant number of features are selected from each feature of the learning patterns, and the capability of a group is obtained (ST3). The learning patterns are divided into ω groups by a clustering method based on the selected features (ST4). This ω is an integer larger than 1. Then, the center value of the ω groups obtained by clustering the learning patterns is applied to the group as a central pattern, overlapping is obtained, and the best grouping is obtained (ST5). The clustering method includes a random clustering, ISODATA clustering, and K-mean clustering, and they are sequentially executed.;COPYRIGHT: (C)1994,JPO
展开▼
