Feature Extraction Methods Comparison between Histogram and PCA in Detecting Occurrences of Stomata on the Freycinetia Sectional Leaves Image

Abstract

Feature extraction is the process of taking an object identifier which can describe the characteristics of the object. In this study, we use two feature extraction methods, namely Histogram and PCA (Principal Component Analysis) to be used in stomata detection from the Freycinetia sectional leaves images. We used a moving frame to detect the occurrence of stomata in an image. To create a classification model, we distinguish between three frame classes: frames showing full stomata, frames showing parts of stomata, and frames which contain no part of stomata at all. For the classification method, we use the Backpropagation Artificial Neural Network for the classifier. The results of the detection process using Histogram as the feature extraction method will be compared with the results of the detection process using PCA. The best results between the two methods of feature extraction will be used as the first step in the process of species type identification for the genus Freycinetia. The research results show that the PCA feature extraction method is better than the Histogram feature extraction method in detecting the occurrence of stomata on the Freycinetia sectional leaves images. The best f1-measure value that can be achieved by the PCA feature extraction methods is 0.9091.