Title:
【発明の名称】パターン認識方法及び装置
Document Type and Number:
Japanese Patent JP3372631
Kind Code:
B2
Abstract:
PURPOSE:To efficiently narrow down the range of categories by converting a feature vector extracted from an input pattern to a vector which is effective for separation between categories and calculating the inner product between this vector and the base vector to select a category group. CONSTITUTION:An input part 11 converts a document picture as the recognition object to an electric signal, and a feature extracting part 12 extracts features peculiar to characters from the electric signal with respect to each character to generate the feature vector. A conversion part 13 converts this feature vector to the feature vector effective for separation between categories, and a rough classification operation part 15 calculates the inner product between this converted vector and the preliminarily determined base vector. A category table reference part 17 refers to a prescribed category table 18 based on the calculation result in the rough classification operation part 15 to select and output the category group to which the input pattern belongs. Thus, the range of categories to which the input vector probably belongs out of many candidate categories is efficiently narrowed down, and pattern recognition is performed with a small calculation volume.
Inventors:
Toru Futaki
Application Number:
JP1637794A
Publication Date:
February 04, 2003
Filing Date:
February 10, 1994
Export Citation:
Assignee:
Canon Inc
International Classes:
G06K9/62; (IPC1-7): G06K9/62
Domestic Patent References:
| JP1267789A | ||||
| JP1267787A | ||||
| JP5197911A | ||||
| JP6251156A | ||||
| JP6417182A | ||||
| JP6273391A |
Other References:
距離関数の学習による手書き数字認識，電子情報通信学会論文誌 Ｊ７６−Ｄ−ＩＩ，日本，１９９３年９月２５日，第Ｊ７６−Ｄ−ＩＩ巻 第９号，ｐｐ．１８５１−１８５９
Attorney, Agent or Firm:
Keizo Nishiyama (1 person outside)