Optimasi Akurasi Algoritma Self PCA Menggunakan Metode CLAHE dan Otsu Thresholding pada Klasifikasi Gender Berdasarkan Citra Wajah

Abstract

In the modern era, the ability of computers to process images has advanced significantly, including facial image recognition for gender classification. This classification can be applied in various contexts, such as product recommendation systems. This study utilizes the Self-PCA algorithm for gender classification based on facial images, employing a separate eigen space approach for male and female classes. Additionally, this research evaluates the effectiveness of preprocessing methods, namely Clahe and Otsu-Thresholding, on three datasets: FGNET, ORL, and Indian, to improve classification accuracy. The data set is randomized and divided into training and testing sets with a ratio of 3:7. Then the data will go through a preprocessing stage, starting with face detection using the Viola-Jones method, applying Clahe and Otsu-Thresholding as preprocessing stages, then training the model with the Self-PCA algorithm and classified using Euclidean Distance. The results demonstrate that the Clahe method increased accuracy for the FGNET dataset 4% and Indian dataset 3.9% but reduced accuracy for the ORL dataset by 4%. Meanwhile, the Otsu-Thresholding method achieved 100% accuracy on the Indian dataset with an 11.1% increase, but decreased accuracy on the FGNET dataset 1.2% and ORL dataset 5%. These findings indicate that dataset quality significantly influences preprocessing effectiveness. Datasets with lighting issues benefited from the Clahe method, while the Otsu-Thresholding method was more effective on datasets with high contrast between objects and the background. In conclusion, the preprocessing stage must be adapted to the characteristics of the data set to optimize classification accuracy.