An Improved Handwritten Javanese Script Recognition using Adaptive Threshold and Multi-Feature Extraction

Abstract

Image quality greatly affects the object recognition process in the image. If the image quality is not good, the recognition process becomes more difficult. Preprocessing, feature extraction, and classifier are the most important parts of the object recognition process in the image. This process will determine object recognition accuracy, precision, and recall. The preprocessing section plays an important role in carrying out a kind of quality improvement so that objects can be easily identified before feature extraction is carried out. This study proposes using an adaptive thresholding method to enhance recognition accuracy in machine learning-based Javanese scripts. The use of adaptive thresholding is carried out in the image binarization process. By using adaptive thresholding, complement, median filter, and dilation operations can be performed to produce a more natural form and pattern of Javanese script writing. Thus, more accurate feature extraction is obtained. Classification is done with the KNN classifier. With a value of K=3, an increase in accuracy of 5% is obtained compared to the previous method.