Image preprocessing analysis in handwritten Javanese character recognition

The handwriting produced by each person is unique, so has a different stroke, even though they write the same letter. Handwritten Javanese an exciting topic to study, in addition scientific purposes and preserving Indonesian culture. character image dataset aksara Jawa: Jawa custom from Kaggle database consists of 2,154 train data 480 evaluation data. This research proposed analyze impact some preprocessing methods recognizing handwritten characters. are dilation, skeletonization, noise reduction. first process segmentation for region interest (ROI) extraction, then various used, finally, recognition step neural network (NN) measure effectiveness method. experiment shows that all (dilation, reduction) give excellent results, especially on black background color, reaching 98% accuracy. Other experimental findings show any combination, accuracy better than white one.