Unconstrained handwritten word recognition with interconnected hidden markov models = 상호 연결된 은닉 마르코프 모델을 이용한 무제약 필기 단어 인식

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 507
  • Download : 0
Recognition of unconstrained handwriting is a challenging task since writers may mix different styles, connecting characters and delaying strokes to continue writing next characters. Several difficulties are anticipated in developing successful recognizers for truly unconstrained handwritings. The first sort of difficulties is due to large shape variations of mixed character styles. The second sort of difficulties lies in the segmentation of cursively handwritten words into individual characters. The third sort of difficulties is due to ligatures, which are formed at the end of character strokes by dragging pen into next character without pen lift. Successful recognizer should be capable to separate character parts from ligature parts to ease the character recognition task. Handling delayed stroke is another difficulty. In this thesis, we proposed an approach for recognizing unconstrained handwritten English words using circular interconnection of hidden Markov models (HMMs). Hidden Markov modeling is the most popular stochastic approach, successfully applied in speech recognition systems and recently being applied for character recognition problems. In order to model words of indefinite length, we interconnected character models circularly with intermediate ligature models. It is based on the view that a handwritten word is an alternating sequence of character and ligature. With the interconnected network of HMMs, the recognition problem is regarded as finding the maximal probability path from the initial state to the final state through HMMs for the given input sequence. From the maximal probability path which consists of an alternating sequence of character HMMs and ligature HMMs, optimal segmentation and associated character labels are obtained simultaneously. Such a network approach allows a natural combination of recognition with post processing utilizing language models. In the word recognition experiments, a lexicon of about 25,000 words was used. For...
Advisors
Kim, Jin-Hyungresearcher김진형researcher
Description
한국과학기술원 : 전산학과,
Publisher
한국과학기술원
Issue Date
1994
Identifier
69085/325007 / 000875457
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전산학과, 1994.2, [ vii, 100 p. ]

Keywords

은닉 마르코프 모델.

URI
http://hdl.handle.net/10203/33013
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=69085&flag=dissertation
Appears in Collection
CS-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0