DocumentCode :
3486006
Title :
Degraded Digit Restoration Based on Physical Forces
Author :
Filho, A. N. G. Lopes ; Mello, C.A.B.
Author_Institution :
Center of Inf., Fed. Univ. of Pernambuco, Recife, Brazil
fYear :
2013
fDate :
25-28 Aug. 2013
Firstpage :
195
Lastpage :
199
Abstract :
It is introduced in this paper an improved method for restoration of degraded digits. The proposed method uses a circular path detection algorithm to determine what type of stroke is being treated. The stroke type will determine which of two techniques will be used to exam the digit´s stroke. The techniques used to analyze the stroke are based on inertial or centripetal forces. The method then artificially re-creates the stroke segments, according to the stroke type, in order to reconstruct the digit. The main objective is the creation of strokes in a way as natural as possible, maintaining the original digit integrity. An artificially created degraded digit dataset was used to conduct the experiments, created based on NIST SD19 digit database. It is shown that the proposed algorithm is successful in its task to reconstruct the digits by artificial stroke extrapolation using information from the digits themselves. The results were analyzed using automatic character recognition and the hit rate for the reconstructed digits presented improvement when compared against the hit rate for degraded digits recognition.
Keywords :
character recognition; image restoration; NIST SD19 digit database; artificial stroke extrapolation; artificially created degraded digit dataset; automatic character recognition; centripetal forces; circular path detection algorithm; degraded digit restoration; digit integrity; hit rate; inertial forces; physical forces; stroke segments re-creation; stroke type; Character recognition; Force; History; Image reconstruction; Image restoration; Image segmentation; Ink; Degraded digits; centripetal force; inertial force; physical forces; restoration;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Document Analysis and Recognition (ICDAR), 2013 12th International Conference on
Conference_Location :
Washington, DC
ISSN :
1520-5363
Type :
conf
DOI :
10.1109/ICDAR.2013.46
Filename :
6628611
Link To Document :
بازگشت