This paper explores a facial recognition system that incorporates signal processing techniques, specifically Fast Fourier Transform (FFT) and Wavelet Transform (WT), to enhance privacy by preventing the reconstruction of original images while maintaining recognition accuracy. The authors propose various filter methods, including conventional and supervised learning filters, to select features effectively, identifying that the Signal-to-Noise Ratio (SNR) and t-test filters yield the best performance. The study demonstrates that filtering and FFT phase removal can successfully protect privacy, achieving minimal impact on recognition accuracy across evaluated datasets.