Comparison between Shape-Based and Area-Based Features Extraction for Java Character Recognition

Java language is one of the local languages are widely used in Indonesia. Java language is widely used by resident of the island of Java. Java language
has special character called Java character. In this research we compare features extraction which will be used to perform the recognition of Java
character. The accuracy of recognition is greatly aected by accuracy of features extraction. Because if there are a lot of similar features between one
character with other characters, may cause the system to recognize as the same characters. In this research, we compare between shape-based features
and area-based features. Shape-based features consist of curves, lines, and loop composing a Java character. The number of curves, lines, and loop will
vary between characters with other characters. For area-based features extraction, each character divide into 9  9 equal regions. In each region, the
number of pixels will be calculated. From experimental results, area-based features extraction gives better result than shape-based features extraction.
This experiment is done by using probabilistic neural network (PNN) as a method of recognition. By using shape-based features extraction, the system
only has recognition accuracy below 20%, but using area-based features extraction, the recognition accuracy can achieve more than 60%.