To distill high-dimensional facial data for classification, I developed a Computer Vision pipeline using PCA to extract structural landmarks from 4,096-dimensional vectors. By implementing StandardScaler and whitening, I ensured that geometric variance, rather than lighting noise, drove the decomposition. I visualized these features as 'eigenfaces' and used RandomizedSearchCV to identify 130 principal components as the optimal signal-to-noise threshold, achieving 81.3% accuracy.