TY  - CONF
PB  - Institute of Electrical and Electronics Engineers Inc.
SN  - 9781479982493
Y1  - 2015///
EP  - 105
A1  - Xiang, W.Y.
A1  - Sebastian, P.
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-84957893479&doi=10.1109%2fCSPA.2015.7225626&partnerID=40&md5=a373adbab412189af38fa62f94e1baab
AV  - none
ID  - scholars5785
TI  - Handwriting recognition using webcam for data entry
SP  - 100
KW  - Backpropagation; Backpropagation algorithms; Character recognition; Data acquisition; Edge detection; Extraction; Feature extraction; Neural networks; Pattern recognition; Signal processing; Software testing
KW  -  Character vectors; Compression modules; Detection modules; Feature extraction methods; Handwriting recognition; Handwritten numeral recognition; Recognition accuracy; Setpoints
KW  -  Image processing
N1  - cited By 0; Conference of IEEE 11th International Colloquium on Signal Processing and Its Applications, CSPA 2015 ; Conference Date: 6 March 2015 Through 8 March 2015; Conference Code:117264
N2  - This paper presents the development of a system that is robust enough to recognize numerical handwritings with the lowest error. The first test was done with a neural network trained with only the Character Vector Module as its feature extraction method. A result that is far below the set point of the recognition accuracy was achieved, a mere average of 64.67 accuracy. However, the testing were later enhanced with another feature extraction module, which consists of the combination of Character Vector Module, Kirsch Edge Detection Module, Alphabet Profile Feature Extraction Module, Modified Character Module and Image Compression Module. The modules have its distinct characteristics which is trained using the Back-Propagation algorithm to cluster the pattern recognition capabilities among different samples of handwriting. Several untrained samples of numerical handwritten data were obtained at random from various people to be tested with the program. The second tests shows far greater results compared to the first test, have yielded an average of 84.52 accuracy. Further feature extraction modules are being recommended and an additional feature extraction module was added for the third test, which successfully yields 90.67. © 2015 IEEE.
ER  -