Implementation and Quantitative Evaluation of Geometric Transformations on Facial Images Using OpenCV Platform

Abstract

This study aims to implement and quantitatively evaluate geometric transformation techniques on facial images using OpenCV in the Google Colab environment. The research was conducted because facial images commonly experience variations in position, orientation, and size during acquisition, which may affect image quality and computer vision performance. An experimental approach was applied using Python, OpenCV, NumPy, Matplotlib, and Google Colab to process facial image data computationally. The implementation stages included image preprocessing, parameter determination, translation, rotation, scaling, result visualization, processing time measurement, and quantitative evaluation using MSE, PSNR, and SSIM. The experimental results showed that translation shifted facial image positions while maintaining output dimensions, rotation changed image orientation but could produce border areas and potential edge information loss, while scaling modified image dimensions proportionally and influenced visual detail according to the scaling factor. Computational testing indicated that all transformation methods were executed efficiently, although stronger validation requires repeated trials and larger datasets. These findings show that geometric transformation techniques can support facial image preprocessing, image simulation, and computer vision development. This study is limited to static images and selected parameters; therefore, future research should include larger datasets, real-time processing, face detection evaluation, and additional transformation methods under more diverse acquisition conditions and scenarios.