Handwriting Recognition for under 5 year-old Children using Artificial Neural Network

• Published in: International Journal of Advanced Studies in Computers, Science and Engineering Vol. 6; no. 8; p. 28
• Main Authors: Wulansari, Zunita, Muslim, M Aziz, Setyowati, Onny
• Format: Journal Article
• Language: English
• Published: Gothenburg International Journal of Advanced Studies in Computers, Science and Engineering 01.01.2017
• Subjects: Artificial neural networks; Back propagation; Braking systems; Children; Children & youth; Fuzzy logic; Handwriting; Handwriting recognition; Image segmentation; Kindergarten; Learning; Methods; Multimedia; Neural networks; Pattern recognition; Pattern recognition systems; Pictures; Preprocessing; Propagation; Software; Students; Studies; Success; Textbooks; Writing
• ISSN: 2278-7917

The linguistic ability components consist of speaking, observing, reading, and writing. For 4 to 5 year-old children, it is easier to learn the ability of reading and writing using multimedia, e.g. pictures, stories, songs, and voices, rather than using textbooks. They need good teachers in learning process. The early capability to write is a main requirement for children to understand subjects given in elementary school. One choice to help improve children's interest and capability in reading and writing could be by utilizing software as the learning media. Such software can then be developed to challenge children with several different levels in order to accelerate their writing ability. The backpropagation method delivers the highest percentage in handwriting recognition among other methods, which makes it as the main option to implement in this research. This method is then used to recognize children's handwriting pattern and then to give scores to them based on the MSE (mean squared error). Furthermore, this method can train multi-layer neuron network so it learns the appropriate internal representation and make it possible to analyze mismatches in input to output mapping. This handwriting recognition process will be done in five sequences. The first stage is preprocessing, which increases image clarity against noise and simplifies the image to make it ready to be analyzed. The second stage is image segmentation, which divide every character into segmented distinct images. The third stage is normalization of each previous segmentation result, and automatically defines position, width, and height of each character. Every segmented character image is read by the network as initial input value. The fourth stage is recognition, which is done by training process in backpropagation neural network previously, and this stage recognizes by comparing its network value with the current one, resulting the squared error. The final stage is the extraction process, which produce the calculation of MSE of entire handwriting sequence. Soon after the handwriting pattern recognized, the software then displays the scores of every child. This application should be used for 4 to 5-year-old children in handwriting learning process. Furthermore, the student ability testing suggests a significant improvement in day-by-day examination, as performed in several kindergartens in Blitar. The success level of backpropagation method to recognize uppercases A to Z is 59.57%, lowercases a to z is 83.0042%, and the arabic numbers 0 to 9 is 75% The result of software implementation has been done for 4 to 5-year-old children in kindergardens for 6 days. It produces 6.63 average score improvement per day.